Control strategies for power distribution networks with ... · EVs into the power distribution...

General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

Users may download and print one copy of any publication from the public portal for the purpose of private study or research.

You may not further distribute the material or use it for any profit-making activity or commercial gain

You may freely distribute the URL identifying the publication in the public portal If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

Downloaded from orbit.dtu.dk on: Apr 06, 2020

Control strategies for powerdistribution networks with electricvehicles integration.

Hu, Junjie

Publication date:2014

Document VersionPublisher's PDF, also known as Version of record

Link back to DTU Orbit

Citation (APA):Hu, J. (2014). Control strategies for power distribution networks with electric vehicles integration. TechnicalUniversity of Denmark, Department of Electrical Engineering. Elektro-PHD

https://orbit.dtu.dk/en/publications/029ca7ad-2e5c-4509-be2f-460da9c32da1

Control strategies for powerdistribution networks with electric

vehicles integration

Junjie Hu

Kongens Lyngby, January 2014ELEKTRO-PHD-2014-01

Control strategies for power distribution networks with electric vehi-cles integration

Author(s):Junjie Hu

Supervisor(s):Prof. Jacob Østergaard, Technical University Of DenmarkProf. Emeritus. Morten Lind, Technical University Of Denmark

PhD school:Department of Electrical Engineering, Technical University of Denmark

Department of Electrical EngineeringCentre for Electric Power and Energy (CEE)Technical University of DenmarkElektrovej 325DK-2800 Kgs. LyngbyDenmark

www.cee.dtu.dkTel: (+45) 45 25 35 00Fax: (+45) 45 88 61 11E-mail: [email protected]

Release date: June 2014Class: 1 (public)Edition: 1Comments: This report is a part of the requirements to achieve

the PhD degree at the Technical University of Den-mark.

Rights: c©DTU Electrical Engineering, 2014

Summary

Demand side resources, like electric vehicles (EVs), can become integral partsof a smart grids because instead of just consuming power they are capable ofproviding valuable services to power systems. EVs can be used to balance theintermittent renewable energy resources such as wind and solar. EVs can absorbenergy during periods of high electricity production and feed the electricity backinto the grid when the demand is high or in situations of insufficient electricitygeneration. However, extra loads created by the increasing number of EVs mayhave adverse impacts on the distribution network such as congestion. Thesefactors will bring new challenges to the distribution system operator. Typically,the challenges are solved by expanding the grid to fit the size and the pattern ofthe demand. As an alternative, the capacity problem can also be solved smartlyusing advanced control strategies supported by an increased use of informationand communication technology. This is the idea of the smart grid. The smartgrid is a next-generation electrical power system that is typified by the increaseduse of communications and information technology in the generation, deliveryand consumption of electrical energy. A smart grid can also be defined asan electricity network that can intelligently integrate the actions of all usersconnected to it - generators, consumers and those that do both - in order toefficiently deliver sustainable, economic and secure electricity supplies.

This thesis focuses on designing control strategies for congestion control in dis-tribution network with multiple actors, such as the distribution system operator(DSO), fleet operators (FO), and electric vehicle owners (or prosumers), con-sidering their self-interests and operational constraints. Note that the controlproblem investigated here deals with “higher level” control, e.g., optimizationstrategy algorithms related scheduling instead of “lower level” direct processcontrol. The thesis starts with reviewing innovative control strategies for largescale management of EVs in the power systems including centralized directcontrol, market based control, and price control. The thesis investigates newapproaches for distribution networks congestion management. It suggests anddevelops a market based control for distribution grid congestion management.The general equilibrium market mechanism is utilized in the operation of the

ii

market.

To build a complete solution for integration of EVs into the distribution net-work, a price coordinated hierarchical scheduling system is proposed which canwell characterize the involved actors in the smart grid. With this system, wedemonstrate that it is possible to schedule the charging scheme of EVs accordingto the users’ energy driving requirements and the forecasted day-ahead electric-ity market price. Several electric vehicle fleet operators are specified to managethe electric vehicle fleets. The method of market based control can then be usedby the DSO to interact with the electric vehicle fleet operators to eliminate thegrid congestion problem. Note that the electric vehicle fleet operator can man-age the EVs based on the three aforementioned control strategies. To test andevaluate the proposed control strategies, multi-agent concepts is used to modelthe price coordinated hierarchical scheduling system. To implement and demon-strate the multi-agent systems, a novel simulation platform has been developedbased on the integration of JACK (a Java based agent-oriented developmentenvironment) and Matlab/Simulink software.

Resume

Elektrisk drevne køretøjer kan udover at forbruge energi ogsa bidrage aktivttil elforsyningen ved balancering af grønne intermitterende energikilder sasomvindmøller og solceller. Elektriske køretøjer kan saledes absorbere energi i pe-rioder med overproduktion og levere energien tilbage til el nettet i situationermed forøget efterspørgsel eller utilstrækkelig produktion. Et øget antal elek-triske køretøjer vil imidlertid ogsa have negativ indflydelse pa elforsyningen,idet de i nogle situationer kan forarsage overbelastning af forsyningslinjer ellerandet udstyr i nettet. Eldistributionsselskaberne vil derfor fa nye udfordringernar antal af elektriske køretøjer forøges. Udvidelser af nettet med flere forsyn-ingslinjer til handtering af den øgede belastning er en af de mulige løsninger.Overbelastning af nettet kan imidlertid ogsa imødegas ved en øget anvendelseaf informations og kommunikationsteknologi (IKT) til intelligent styring. Detteer ideen bag udviklingen af “smart-grid”, som er fremtidens elforsynings sys-tem karakteriseret ved en øget anvendelse af IKT i hele forsyningskæden fraproduktion, leverance og forbrug af elektrisk energi.

Afhandlingen emne er udvikling af styringsstrategier for handtering af overbe-lastning i eldistributions net med mange aktører sasom netoperatører, fladeoper-atører samt ejere af elektriske køretøjer. Styrings-strategierne udformes saledesat de tager hensyn til aktørernes interesser samt nettets driftsbegrænsninger.Afhandlingen fokuserer pa overordnede strategier til optimering af driftsplaner(schedules) for nettet og de forbundne køretøjer. Styring af de enkelte køretøjerer saledes ikke behandlet.

Afhandlingen indledes med en oversigt over innovative strategier for overord-net handtering af elektriske køretøjer i elforsyningen omfattende centraliseretdirekte styring samt markeds- og prisbaseret styring. Derefter udvikles enstrategi for markeds-baseret styring af distributionsnettet som handterer over-belastninger (congestion) baseret pa “general equilibrium market” mekanismer.Derudover foreslas en prisbaseret hierarkisk styring til integration af elektriskekørertøjer i distributionsnettets drift, som muliggør inddragelse af aktørernes in-teresser. Det demonstreres at nar disse to styringsprincipper kombineres er detmuligt at schedulere opladningen af elektriske køretøjer under hensyntagen til

iv

brugernes behov for energi til transport samt dagsprisen for elektricitet pa day-ahead markedet. Det udviklede styringsprincip understøtter mere end en fleetoperatør og markedsbaseringen gør det muligt for netoperatøren at interageremed dem ved handteringen af overbelastningssituationer i eldistributionsnettet.Fladeoperatører kan tillige anvende centraliseret direkte styring, markedsbasereteller prisbaseret styring af de enkelte elektriske køretøjer. De udviklede strate-gier demonstreres og evalueres ved en implementering baseret pa en integrationaf multiagent software teknologi (JACK) samt Matlab/Simulink software.

Preface

This thesis was prepared at Center for Electric Power and Energy, Departmentof Electrical Engineering, Technical University of Denmark in partial fulfillmentof the requirements for acquiring the Ph.D. degree in engineering.

This thesis deals with developing control strategies for large scale integration ofEVs into the power distribution network. It mainly addresses applications ofmarket based control strategy for cost efficient and flexible control of distributedelectric vehicles within the distribution networks. The research questions arederived from the state of the art in the electric power systems and the newchallenges and requirements faced by the power industry. The project has beenmainly supported by and connected to two Danish research and innovationprojects, the Edision project 1 and the iPower project 2. In the beginning ofthe PhD study, I was mainly working on the subject of electric vehicle smartcharging, learning from the discussions and results generated from the Edisonproject (the project was finished around the middle of 2011). Then, gradually,my working time and attention were moving to the subject which is investigatedin the iPower project (mainly work package 3, 4), i.e., distribution grid conges-tion related research. An outcome of this work package is the FLECH (flexibilityclearing house) platform. In addition, the PhD student would like to thanks thesupport from the Ph.D. Programs Foundation of Ministry of Education of Chinaunder the grant (20100072110038).

In general, the research topics of the PhD project are inspired by the discussionfrom the Edison and iPower project, but the method developments such asmarket based control for distribution grid congestion management are almostindependent from the iPower project. Furthermore, we took the decision tomodel and evaluate the market based control approach developed for solvingdistribution grid congestion in a multi-agent system. Currently, our multi-agentsystems based platform has been proposed and preliminarily used to support theFLECH mock up platform simulation which is more flexible and can simulatevarious services required by the distribution system operator.

1http://www.edison-net.dk/2http://www.ipower-net.dk/

vi

This thesis consists of a summary report and a collection of nine research paperswritten during the period 2011–2013.

Acknowledgements

Firstly, I would like to express my sincere gratitude to my supervisor ProfessorEmeritus Morten Lind for his kind supervision, patience and mentoring duringthe course of my PhD study. He has always been a source of guidance andinspiration during the research and writing of this thesis as well as on philosophyof science, engineering and life in general. I would also like to express myheartfelt gratitude to my supervisor Prof. Jacob Østergaard for his motivation,instructive supervision and guidance during the course of my PhD study. Hehas been an example of pursing excellence during my study and will always bean example for my future career.

I am especially thankful to my colleague Dr. Shi You for his generous discussionand guidance through my PhD study as well as his help which let me adapt tothe life in Denmark smoothly. He is always willing to share his knowledge and itis a pleasure to work together with him. I am especially thankful to my externaladvisors Prof Lars Nordstrom and Dr. Arshad Saleem for their kind supervi-sion and instructive discussions. They provided a nice working environmentduring my four months external stay at Royal Institute of Technology (KTH,Sweden). It is really a pleasure to work there. I am also grateful to PhD studentNicholas Honeth from Royal Institute of Technology for his help on the JACKimplementation.

I am greatly thankful to Dr. Hugo Morais for the interesting joint work andfor his proof-reading of this thesis which brought some instructive technicaldiscussions. I am also thankful to my colleague assistant Prof. Kai Heussen forhis kind discussion and generosity of sharing his knowledge. It is a pleasure tocollaborate with him and being an teaching assistant to his course IntelligentSystems. I am greatly thankful to the management and administrational teamincluding my group leader senior scientist Henrik W. Bindner, center secretaryLouise Busch-Jensen, Eva M. Nielsen, and technical assistant Georg Hvirgeltoft.Your consistent support, guidance and provision of all required resources meantthat I could concentrate on my work with full peace of mind and concentration.

viii

I am thankful to all my colleagues at DTU, especially Dr. Peter Bach Andersen,Ph.D. student Xue Han, Danel Esteban Bondy, Anders Bro Pedersen for theinteresting joint work and thoughtful discussion, PhD student Philip JamesDouglass, Tilman Weckesser, Haoran Zhao, Zhaoxi Liu, Seyedmostafa HashemiToghroljerdi for sharing many interesting discussion on various topics related toour common interests of research and study.

I am also thankful to my team members at iPower project WP4 especially LarsHenrik Hansen from DONG Energy and Olle Sundstrom at IBM, Zurich forintroducing me to several interesting challenges in the industry and providinginputs. I am also thankful to Benjamin Biegel from Aalborg University for theinteresting discussion.

I would like to express my gratitude to my teachers and friends in China, theyare Prof. Lei Wang from TongJi University, associated Prof. Bin Li, Prof. HongChen, associated Prof. Hongru Tang from Yangzhou University for their friendlytreating and encouragement when I was back in China during the Christmasholidays. Besides, I am greatly thankful to my fellow classmates PhD studentTian Lan from Technical University of Berlin, Dr. Chengyong Si from Univer-sity of Shanghai for Science and Technology, Dr. Weian Guo, Mr. ChengyuHuang from TongJi University for their discussion on optimization problemsand interesting joint work.

Finally and most importantly, I am greatly thankful to my family. Your love,care and attention make my life worth living, struggling and progressing. Loveyou all.

x Contents

Contents

Summary i

Resume iii

Preface v

Acknowledgements vii

Contents xiii

Glossary xv

1 Introduction 1

1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Objectives and research problems . . . . . . . . . . . . . . . . . . 2

1.2.1 Overall objectives . . . . . . . . . . . . . . . . . . . . . . 2

1.2.2 Problems analysis . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Research contributions . . . . . . . . . . . . . . . . . . . . . . . . 6

1.4 Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.5 Outline of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 State of the art 11

2.1 Management of electric vehicle fleet and their business opportunities 11

2.2 Control strategies for distribution grid congestion management . 16

2.3 Multi-agents systems for smart distribution grid operation andsimulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

xii CONTENTS

3 Optimization and control for integrating electric vehicles 253.1 Background knowledge . . . . . . . . . . . . . . . . . . . . . . . . 25

3.1.1 Nordic electricity market . . . . . . . . . . . . . . . . . . 253.1.2 Battery modeling . . . . . . . . . . . . . . . . . . . . . . . 263.1.3 Driving pattern and associated electricity consumption . . 28

3.2 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293.2.1 Three control strategies for integrating electric vehicles . 293.2.2 Formulation of the optimal charging of electric vehicles by

centralized control architecture . . . . . . . . . . . . . . . 313.2.3 Formulation of the optimal charging of electric vehicles

using decentralized control architecture . . . . . . . . . . 363.3 Summary and discussions . . . . . . . . . . . . . . . . . . . . . . 37

4 Market based control for distribution grid congestion manage-ment 394.1 Formulation of the congestion problems in distribution network . 39

4.1.1 Electricity consumption by electric vehicles and its impacton the distribution grids . . . . . . . . . . . . . . . . . . . 40

4.1.2 Congestions management in distribution network opera-tion and its enabling functions . . . . . . . . . . . . . . . 41

4.2 Background knowledge of the market based control solution . . . 424.2.1 What is market based control? . . . . . . . . . . . . . . . 424.2.2 General equilibrium market mechanisms . . . . . . . . . . 43

4.3 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444.3.1 Map of the market based control operation . . . . . . . . 444.3.2 Market based approach for grid congestion prevention . . 454.3.3 Price coordinated hierarchical scheduling system . . . . . 454.3.4 A flexibility clearinghouse concept (FLECH) . . . . . . . 47

4.4 Summary and discussions . . . . . . . . . . . . . . . . . . . . . . 484.4.1 Comparisons among the three control strategies . . . . . . 484.4.2 Market based control related issue . . . . . . . . . . . . . 51

5 Multi-agent system for distribution grid congestion manage-ment 535.1 Characteristics of multi-agents system . . . . . . . . . . . . . . . 535.2 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

5.2.1 Modeling the price coordinated hierarchical schedulingsystem by multi-agent concepts . . . . . . . . . . . . . . . 56

5.2.2 Platforms Development . . . . . . . . . . . . . . . . . . . 575.3 Summary and discussions . . . . . . . . . . . . . . . . . . . . . . 59

6 Conclusions and Future Work 616.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 616.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

CONTENTS xiii

Bibliography 65

Appendices 75

A Core publications 75A.1 Optimization and control methods for smart charging of electric

vehicles facilitated by fleet operator: review and classification . . 75A.2 Optimal charging schedule of an electric vehicle fleet . . . . . . . 103A.3 Optimal control of an electric vehicle’s charging schedule under

electricity markets . . . . . . . . . . . . . . . . . . . . . . . . . . 110A.4 Numerical comparison of optimal charging schemes for electric

vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119A.5 Coordination strategies for distribution grid congestion manage-

ment in a multi-actor, multi-objective setting . . . . . . . . . . . 126A.6 Coordinated charging of electric vehicles for congestion preven-

tion in the distribution grid . . . . . . . . . . . . . . . . . . . . . 135A.7 Multilevel coordination in smart grids for congestion management

of distribution grid . . . . . . . . . . . . . . . . . . . . . . . . . . 145A.8 A clearinghouse concept for distribution-level flexibility services . 152A.9 A multi-agent system for distribution grid congestion manage-

ment with electric vehicles . . . . . . . . . . . . . . . . . . . . . . 158A.10 HV/MV power transformer capacity negotiation by fleet opera-

tors using multi-agents . . . . . . . . . . . . . . . . . . . . . . . . 170

B Supplementary background and methods 183B.1 Battery modeling-Continuous . . . . . . . . . . . . . . . . . . . . 183B.2 Price control for regulating electric vehicles’ charging behavior . 185B.3 Congestion management principles in transmission systems . . . 186B.4 Contract net protocols and auction approach . . . . . . . . . . . 189B.5 Decomposition methods and subgradient method . . . . . . . . . 190B.6 Market based control-Negotiation types . . . . . . . . . . . . . . 193B.7 Alternating direction method of multipliers (ADMM) . . . . . . . 193

xiv CONTENTS

Glossary

Battery pack

A battery pack is the final assembly used to store and discharge electric energyin a EV.

Battery State of charge (SOC)

The available capacity in a battery expressed as a percentage of rated nominalcapacity.

Centralized direct control

The up-level controller directly schedule and specify the low-level units to exe-cute the commands. It can be applied to both the control relations between theDSO and the fleet operators in term of grid capacity allocations and between thefleet operators and the individual electric vehicles in term of charging operation.

Demand side resources (DSR)

DSR refers to the geographically distributed modular power generation, con-sumption and energy storage systems which are located on the demand sideand have the capability of altering their consumption pattern.

Distributed energy resources (DER)

A power producing or consuming unit connected to a distribution system withcommunication and explicit control capabilities that could be employed to sup-port the electricity grid.

Distribution system operator (DSO)

A system operator responsible for distribution systems.

xvi

Electric vehicle (EV)

EV is defined as passenger vehicle exclusively with an electric drive.

Electric vehicle fleet operator (EV FO)

EV FO is used to manage electric vehicle and represent them to interact withthe market. EV FO could be independent or integrated in an existing businessfunction of the energy supplier. Simply, in this thesis, FO means EV FO.

Flexibility clearing house (FLECH)

A platform developed in the Danish iPower project (http://www.ipower-net.dk/)for facilitating ancillary services at the distribution system level.

Market based control

Market based control is a paradigm for controlling complex systems with con-flicting resources. It typically includes the features found in a market such asdecentralized decision making and interacting agents. Market based controlusually requires two way communication, e.g., exchange the price and powerschedule information. It can be applied to both the control relations betweenthe DSO and the fleet operators in term of grid capacity allocations and be-tween the fleet operators and the individual electric vehicles in term of chargingoperation.

Price control

Price control applies broadcasting of price signal with a regular update frequencyto the demand side resource. Price control is in the form of one way commu-nication. It can be applied to both the control relations between the DSO andthe fleet operators in term of grid capacity allocations and between the fleetoperators and the individual electric vehicles in term of charging operation.

Transmission system operator (TSO)

A System Operator responsible for the transmission system.

Vehicle to grid (V2G)

V2G describes a system in which plug-in electric vehicles connect to the powergrid to deliver the electricity into the grid.

Chapter 1

Introduction

1.1 Background

An important means used by the power industry to reduce greenhouse gas emis-sion and fossil fuel dependency is the introduction of renewable energy resources,such as wind and solar generation. Denmark was a pioneer in wind power whichprovides a large amount of electricity to consumers. By the end of 2012, the to-tal installed wind capacity in the Danish power systems was 4162 MW counting30.1% of the domestic electricity usage [1]. Wind energy in Denmark is expectedto grow due to the political strategy of achieving 50% wind power in the 2020Danish power system [2]. The most installed wind power in Denmark is con-nected at the distribution system level, which brings challenges to Energinet.dk(transmission system operator (TSO) of Denmark). Energinet.dk has limited orno access to the information about the status at the medium grid voltage level.In order to address the challenges, several actions [3] have been implemented orplanned, such as:

• Coordination of the power flows among different systems by electrical in-terconnections, mostly high voltage direct current to the TSOs in theSweden, Norway, Germany, and soon the Netherland.

• Balancing the power systems by the deregulated power market with thecollaborations of power balance responsible parties. The balance respon-sible parties make the power and energy bids into the market, consistingof conventional power and wind power.

• Implementing tools to provide real time estimation of the amount of powerinjections from wind energy.

2 Introduction

• Managing the flexible demand, like electric vehicles (EVs), heat pumps inthe residential area, large industrial factory and commercial buildings etc.

In order to manage the flexible demand, such as electric vehicles’ flexibilities,a research project in Denmark with international collaboration named Edisonproject 1 was funded to develop optimal system solutions for electric vehiclessystem integration, including EV battery technology, networks issues, marketsolutions, information and communication technology standard development.To utilize the full benefit of the interaction between the electric vehicles andthe power grid with a large amount of power from fluctuating sources, softwaresthat enables electric vehicles to charge when there is a surplus of energy inthe system or to resupply energy to the grid when there is a lack of power inthe system are developed. The major control concept in the Edison projectis the introduction of a fleet operator (FO) to aggregate the consumption ofa number of electric vehicles and handle their interaction with the electricitymarket as one unit with a centralized/direct control [4–6] to capture the fullbenefits. Insights from the project indicated that distribution grid congestionmay happen with the utilization of smart charging scheme [7–9]. In order tosolve the grid congestion caused by the newly increasing demand, well definedcontrol strategies are required for the power distribution systems and this is themain research topic of this PhD project.

1.2 Objectives and research problems

1.2.1 Overall objectives

This thesis deals with development of control strategies for large scale integra-tion of electric vehicles into the power distribution network. Fig. 1.1 illustratesthe relevant actors, operations, and the available control strategies in the con-sidered systems. We hypothesize that it is possible to schedule the chargingscheme of electric vehicles according to the users’ energy driving requirementsand the forecasted day ahead electricity market price. Several fleet operatorsare specified to manage the EV fleets, then distribution system operator (DSO)can interact with the fleet operators to eliminate the grid congestions. Notethat the dispatch currently used is defined only based on the spot market andthe regulation market. The state of the distribution grid is not considered. Weaim to develop a control strategy for distribution grid congestion managementbefore the resource dispatch. Our objectives are to coordinate the self interests

1http://www.edison-net.dk/

1.2 Objectives and research problems 3

and operation constraints of three types of actors in the market: the EV owner,the fleet operators and the distribution system operator facilitated by the de-veloped control strategies, e.g., considering the individual EV owner’s drivingrequirement, the charging cost of electric vehicle, the fleet operator’s businessobjectives and the thermal limits of cables and transformers of distribution grid.

Electricity Spot

Market

Regulation

market

Status

information

Control/

coordination

relation

Physical

connection

Fleet

operation

Distribution

system

operation

EV

operation

Coordination strategies:

Centralized control


Price control

Control strategies:

Centralized control


Price control

Figure 1.1: Power system with electric vehicles integration coordinated by fleetoperators

1.2.2 Problems analysis

In the deregulated electric power industry, the system described in Fig. 1.1 canbe regarded as a typical decentralized, hierarchically organized system. Thesubstantial characteristic of such system is their decomposability into a seriesof individual functional levels [10]. The functional levels of the system usuallyinclude: Level 4, plant management, Level 3, production scheduling, Level 2,plant supervisory control, and Level 1, direct process control. At each level,some automation functions are implemented to operate on the next ”lower”level. The execution of function is, however, initiated and controlled by thenext ”higher level”. With this four-stage system, the structural and functionalproperties of a hierarchical system can be explained. The control problem in-vestigated in this thesis is mainly referred to Level 3 production scheduling.

4 Introduction

This level is responsible for a series of functions, belonging more to the areaof operations research and resource allocation than to the systems or controlengineering. Production scheduling/dispatching for the system according to thestatus of consumer’s requirements, network constraints, and energy demands isthe main concern of this level. At this level, the current existing control meth-ods which are proposed to integrate large scale of distributed energy resourcesinto the power systems including centralized control, market based and pricecontrol 2. Fig. 1.2 overviews the control methods. Note that the local controlmethod is considered, although it is effective and simple to be implemented. Inaddition to this, the local control can be put inside the price control category,for example, the units can set up a command that the device will be turned onif the price is lower than a threshold value.

Centralized

control

Market based

control

Price

control

Communication

form and cost

Computational

requirementsFeatures

Control signals (i.e.,

set points)

High level controller

makes the decision


generates and sends the

price to the units

Low level units respond

with power schedule

Multiple iterations

Privacy improved


generates and sends the

price to the units

Low level units need not to

explicitly respond.

Privacy improved

Low for Control

object

High for controller

Relative high for

Control object

Low for controller

Low for Control

object

Relative high for

controller

One/Two way

High

Two way

Relative high

One way

Low

Figure 1.2: Control strategies overview

Given the three control strategies, the overall problems of developing controlstrategies for large scale integration of electric vehicles into the power distribu-tion network have been decomposed into three subproblems:

1. In order to manage a large scale of electric vehicles, from a commercial

2In addition to the centralized control adopted in the Edison project, market based controland price control are also proposed by other research, more details will be explained in nextChapter regarding state of the art.

1.2 Objectives and research problems 5

actor (fleet operator) or system operator’s 3 perspective, what are the im-plications, advantages and disadvantages when choosing a control strat-egy? Furthermore, how to design the control algorithms and what are thenecessary information and communications that are required to supportthe chosen control strategy?

2. To address the congestion issue of distribution network, with a conditionthat we can benefit more by using the advocated three control strategiesexcept the traditional way of upgrading the grid, what are the implica-tions, advantages and disadvantages for the system operator, the com-mercial players, and the EV owners when choosing a control strategy? Inaddition, how to devise the control algorithms and what are the necessaryinformation and communications that are required to support the chosencontrol strategy?

3. Development of tools to simulate the control strategies developed for dis-tribution grids congestion management with electric vehicle integrationconsidering variously involved actors and different domains.

These three subproblems has been turned into three research topics of thisproject which will be elaborated in below. The three research topics are:

1. Optimization and control for integrating electric vehicles.

The thesis starts to review and compare the control strategies for inte-grating the increasing number of electric vehicles into the power systems.Then, it investigates control algorithms which support the various controlstrategies such as linear programming, dynamic programming and mixedinteger linear programming based techniques.

2. Market based control for distribution gird congestion management.

In this research topic, the market based control is used to prevent thegrid congestion. The focus has been divided into two perspectives: oneis to find the suitable price clearing algorithms for the actors inside themarket taking into account the effectiveness, the computation cost, andthe generality; another one is to find the suitable optimization techniqueswhich support the actors’ participation in the market. Note that themarket based control out of the three control methods is chosen, mainlydue to the following reasons: 1) It fits with the deregulated electricitymarket environment. 2) The congestion management in the transmissionsystems is managed according the market based control method. 3) The

3Note that the commercial actor might not exists if the business opportunity is not attrac-tive. Under the circumstances, the DSO needs to control or coordinate the charging behaviorsof EVs to avoid the grid congestion.

6 Introduction

introduction of the FOs make it possible to manage the congestion in amarket based approach. 4) The certainty of the problem solving is higherthan the price control and the computation requirements for the systemoperator is lower than the centralized control.

3. Multi-agent system for distribution grid congestion management

The research focuses on develop modeling framework and tools for distri-bution grid congestion management study. Precisely, multi-agents systemis used for modeling and evaluation of the developed control strategiesfor distribution grid congestion management. The aim is to prove thatmulti-agent system is a suitable technology to fulfill the requirements.

1.3 Research contributions

The main contributions of the research consist of:

• We present a comprehensive review of the optimization and control meth-ods that are used in integrating electric vehicles into the power systems.It outlines the advantages and disadvantages of various control strategies,presents the details of the modeling method and algorithms in each controlstrategy. Details are presented in separate papers A.1, A.2, A.3, and A.4and Chapter 3 gives a summary of the main results.

• A price coordinated hierarchical scheduling/control system has been devel-oped and demonstrated successfully for the integration of the distributedenergy resources into the power distribution networks. In the system, weapply a market based control strategy to solve the distribution grid con-gestion. The price clearing algorithm used in this market falls into thegeneral equilibrium market scheme. We use the electric vehicles as case toillustrate the systematic integration of distributed energy resources intothe distribution network. With the case, we also demonstrate that var-ious control methods used by the fleet operator such as direct control,price control can be flexibly integrated into this hierarchical control sys-tems. Details are presented in separate papers A.5, A.6, A.7, and A.8 andChapter 4 gives a summary of the main results.

• A multi-agent technology has been motivated and used to simulate/demonstratethe proposed market scheme. We present a flexible and powerful simula-tion platform which is based on the integration of an agent based sim-ulation tool JACK, computational tool Matlab, and grid simulation in

1.4 Publications 7

Simulink. The developed platform can support the simulation of intelli-gent control scheme for the smart grids. Details are presented in separatepaper A.9, A.10 and Chapter 5 gives a summary of the main results.

1.4 Publications

We present these contributions by providing both a summary report as the maincontent of this thesis and a number of papers that we have written throughoutthe project period attached in the appendix. Totally 10 papers are includedin appendices A.1 through A.10. The papers are referenced throughout thisreport as needed, but may also be read independently of this report. Papers A.1to A.10 contain the main results of the project, and are included as appendices.Papers from 11 onwards contain other contributions from the author that arenot directly related to the PhD thesis.

A.1: J. Hu, S. You, C. Si, M. Lind, and J. Østergaard, Optimization and con-trol methods for smart charging of electric vehicles facilitated by fleet operator:review and classification. International journal of distributed energy resourcesand smart grids, 2014.

A.2: J. Hu, S. You, M. Lind, J. Østergaard, and Q.Wu, Optimal charging sched-ule of an electric vehicle fleet, in proceedings of 46th International Universities’power engineering conference (UPEC), 2011, Germany.

A.3: T. Lan, J. Hu, Q. Kang, C. Si, L. Wang, Q. Wu, Optimal control of anelectric vehicle’s charging schedule under electricity markets, Neural computingand applications, 2012.

A.4: S. You, J. Hu et al., Numerical comparison of optimal charging schemesfor electric vehicles, in proceedings of IEEE PES general meeting, San Diego,U.S., 2012.

A.5: P. B. Andersen, J. Hu, K. Heussen, Coordination strategies for distribu-tion grid congestion management in a multi-actor, multi-objective setting, inproceedings of IEEE Innovative Smart Grid Technologies (ISGT) Berlin, 2012.

A.6: J. Hu, S. You, M. Lind, and J. Østergaard, Coordinated charging of electricvehicles for congestion prevention in the distribution grid, To appear in IEEEtransactions on smart grids, 2014.

A.7: J. Hu, M. Lind, Saleem, A, S. You, and J. Østergaard, Multilevel coor-

8 Introduction

dination in smart grids for congestion management of distribution grid, in pro-ceedings of International conference on intelligent system applications to powersystems (ISAP), 2013, Japan.

A.8: K. Heussen, D. E. Bondy, J. Hu, et al., A clearinghouse concept fordistribution-level flexibility services, in proceedings of IEEE Innovative SmartGrid Technologies (ISGT) Copenhagen, 2013.

A.9: J. Hu, A. Saleem, S. You, L. Nordstrom, M. lind, J. Østergaard, A multi-agent system for distribution grid congestion management with electric vehicles,Engineering Application of Artificial Intelligence, in submission, 2013.

A.10: J. Hu, H. Morais, M. lind, J. Østergaard, HV/MV power transformercapacity negotiation by fleet operators using multi-agents, submitted to 12thInternational Conference on Practical Applications of Agents and Multi-AgentSystems, 2014.

11: J. Hu, M. Lind, S. You, X. Zhang, Multilevel flow modeling of domesticheating systems, proceedings of 2012 Symposium on Socially and TechnicallySymbiotic Systems (STSS), Japan.

12: W. Guo, D. Yang, J. Hu, C. Huang, L. Wang and Q. Wu, Optimal manage-ment of a home smart grid scheduling, Journal of Computational InformationSystems, v 8, n 5, p 1921-1928, March 1, 2012.

13: S You, J. Hu, et al., Analytical framework for market-oriented DSR flexi-bility integration and management, Energy and Power Engineering, ISSN Print:1949-243X, 2012.

14: C. Si, T. Lan, J. Hu, L. Wang, Q. Wu, On the penalty parameter of thepenalty function method, to appear in the journal Control and Decision, 2014.(In Chinese)

15: S. You, H. W. Bindner, J. Hu, An Overview of trends in distribution networkplanning: A Movement Towards Smart Planning, accepted for publication inIEEE PES, Transmission and distribution conference, 2014.

Table 1.1 summarizes contribution of the research paper to the three researchtopics of the PhD project.

In more details, for the first research topic, paper A.1 reviews the three controlstrategies for smart charging of electric vehicles. It outlines the informationflows and presents the widely proposed control algorithms in the three differ-ent control strategies. The next three papers focus on methods development,

1.4 Publications 9

Research topics PublicationA.1 A.2 A.3 A.4 A.5 A.6 A.7 A.8 A.9 A.10

1. Optimization and con-trol for integrating electricvehicles

x x x x

2. Market based control fordistribution gird congestionmanagement.

x x x x

3. Multiagent system fordistribution grid congestionmanagement

x x

Table 1.1: Relation between research topics and publications

i.e., using linear approximation and nonlinear approximation to characterize thebattery charging issue and a formulation of the charging problem in a linear pro-gramming based charging cost minimization problem (paper A.2) and dynamicprogramming based charging cost minimization problem (paper A.3). Further-more, vehicle to grid study is performed in paper A.4 to perform the economicanalysis considering its profits of providing regulation services and its batterydegradation costs, mixed integer linear programming based formulation is usedin this paper.

For the second research topic, paper A.5 addresses the interactions betweenthe stake-holders involved in the distribution system (the distribution systemoperator, the commercial actors, and the EV owners) when handling the dis-tribution grid congestion problem, and identifies several approaches by whichtheir diverse and potentially conflicting objectives can be coordinated. PaperA.6 recommends and tests the market based control scheme identified in paperA.5 with detailed mathematical explanation, proof and illustrations with casestudies for congestion prevention. Paper A.7 extends the study in paper A.6and proposes the price coordinated hierarchical control system to control thedistribution grid congestion considering electric vehicle integration. Paper A.8introduces the flexibility clearinghouse (FLECH) concept and presents a casestudy of using electric vehicles to provide services to the distribution systemoperator.

For the third research topic, paper A.9 extends the concepts in paper A.6 andthe extension mainly utilize the multi-agents system (MAS) technology to as-sess the proposed market scheme with the purpose of illustrating and trackingthe coordination behaviors. Moreover, the extension considers demonstratingEVs’ flexibility (through the presence of response weighting factor to the shadowprice) in the developed MAS system. Paper A.10 uses the multi-agents based

10

platform (developed in paper A.9) to simulate a collaborative but also compet-itive environment where multiple fleet operators negotiate on the transformercapacity of the distribution grid. In the study, each fleet operator do its EVscharging/discharging scheduling taking into account the network constraints.

1.5 Outline of the thesis

This thesis comprises 6 chapters. Chapter 1 introduces the overall problemand the research objectives of the project. Chapter 2 provides a review on thestate of the art of the three sub-problems. In chapter 3, the electric vehiclesintegration research is presented in a compact way. In chapter 4, the developedmarket based control strategy is presented. Chapter 5 discusses the topic ofmulti-agent based modeling and simulation. Finally, in chapter 6, conclusionsand future work are presented.

Chapter 2

State of the art

In this chapter, existing literature and practices relevant to the three researchproblems of this PhD project are reviewed. First, in sec. 2.1, we review the po-tential business opportunities associated to electric vehicle fleet and the controlstrategies which are proposed to manage the electric vehicles. Second, in sec.2.2, control strategies for solving the distribution grid congestion problem havebeen reviewed, especially, the review focuses on the methods which are pro-posed for the control relations between the distribution system operation andelectric vehicle fleet operations. Finally, in sec. 2.3, the multiagent based sys-tem and their application in smart distribution grid operation and simulationsare discussed. Note that as each paper attached in the appendices contains theliterature studies and references relevant to the specific topic, the reference listin this chapter is not exhaustive, and the reader is referred to those given inthe papers as well. Nevertheless, the aim is to include and present some studieswith figures which facilitate further understanding.

2.1 Management of electric vehicle fleet and theirbusiness opportunities

Much research has been focused on electric vehicles integration into power sys-tem since the last decades. In this section, we review the publications basedon its control objectives which are summarized from the existing studies, i.e.,start with the discussion on how electric vehicles can be aggregated to provideancillary service to transmission system operator; then present how electric ve-hicles can be used to maximize the production of renewable energy producers;

12 State of the art

next, the discussion will be moved on to the proposed commercial actor, i.e.,fleet operator; finally, the optimization methods of minimizing the charging costof electric vehicles are examined.

Kempton et al. [11–14] analysed the potential profits of vehicle to grid (V2G)support by comparing it to existing ancillary services and found that partici-pating regulation power market appears to be most promising and offers a sub-stantial earning potential to EV owners. This is because: (a) it has the highestmarket value for V2G among the different forms of electric power (much higherthan peak power, for example), (b) it minimally stresses the vehicle power stor-age system, and (c) battery-electric vehicles are especially well suited to provideregulation services.

ISO

Figure 2.1: Illustrative schematic of proposed power line and wireless controlconnections between vehicles and the electric power grid, adapted from [12]

As illustrated in Fig. 2.1, the electric vehicles can participate in the regulationservices individually or by joining a fleet, the communication can be facilitatedby power line and wireless control connections. It is advocated that fleets aremore easily accommodated within existing electric market rules, which typicallyrequire power blocks of 1 MW. To fulfill the concept of V2G, each vehicle musthave three required elements: (a) a connection to the grid for electrical energyflow, (b) control or logical connection necessary for communication with the gridoperator, and (c) controls and metering on-board the vehicle. Fig. 2.2 shows

2.1 Management of electric vehicle fleet and their business opportunities13

an example of electric vehicle control panel. By predefining the wanted drivingdistance and the comfortable buffer, the electric vehicles can be connected tothe grid and then participate the regulation service market.

AUTO CHANGE CONTROLLER

COST SINCE LAST RESET

(NEGATIVE INDICATES CREDIT)

ALWAYS MAINTAIN ENOUGH FOR

Nerve Sell

50 miles

10 miles

2 miles

1 miles

100 miles

50 miles

10 miles

2 miles

1 miles

100 miles

NEXT TRIPDISTANCE NEEDEDFOR NEXT TRIP

TIME OF NEXT TRIP

CHARGEDENOUGH FOR

SET

HOUR

MIN

MILES

NOTE: CHARGE MILEAGI IS ESTIMATED. IT WILL BE IESSFOR FULL LOADS OR HILLY

TERRAIN

CHARGING

DISCHARGING

SELLING ELECTRICITY

6:45AM

20.6

$-15.73

Figure 2.2: Example control panel for a battery electric vehicle, allowing thevehicle operator to constrain charging and discharging by the electric utility,adapted from [11]

It is also learned from the studies [11–14] that important variables for the V2Gmarket are: (a) the value of ancillary services in the area, (b) the power capacityof the electrical connections and wiring, and (c) the kWh capacity of the vehiclebattery. The amount of time the vehicles were on the road or discharged didnot turn out to be a major variable. The results showed that battery electricvehicles fleets have significant potential revenue streams from vehicle to grid.

Regarding the Danish Power system, Divya et al. [15] carried out a study inves-tigating the feasibility of integrating electric vehicles in the Danish electricitynetwork which is characterized by high wind power penetration. They foundthat electric vehicles have the potential to assist in integrating more wind powerin 2025 when the electric vehicle penetration levels would be significant enoughto have an impact on the power systems. Østergaard et al [16–18] shows thatintelligent integration of electric vehicles in the Danish power system with highwind power penetration has substantial socio-economic benefits due to its bal-ancing capability.

14 State of the art

The studies presented in [12–19] are mainly investigated from the design andanalysis perspectives, they gave a good incentive for the electric vehicles toparticipate into the regulation market. The subsequent work focused on themethod development. Rotering and Ilic [20] utilized the dynamical program-ming to formulate the smart charging problem. They took into account vehicleto grid as a mean of generating additional profits by participating in the ancil-lary service markets. Based on the data of the independent system operator ofCalifornia, provision of regulating power substantially improves plug-in hybridelectric vehicle economics and the daily profits amount to 1.71$, including thecost of driving. Han et al. [21] proposed a fleet operator that manages electricvehicles to provide frequency regulation services, the cost arising from the bat-tery charging and the revenue obtained during the participation is investigated.The question is formulated as an optimization problem and dynamic program-ming is used to generate the charging profile. Pillai and Jensen [22] investigatedthe V2G regulation capabilities in the West Denmark power system by using asimplified load frequency control model, in the study, they used an aggregatedbattery storage model and generators model. The results indicated the reg-ulation needs from conventional generators are significantly minimized by thefaster up and down regulation characteristics of the EV battery storage. Allthese results indicate that it is feasible to participate in the electricity marketand provide ancillary service to the gird.

In addition to the focus on providing ancillary services to transmission systemoperator, several studies investigated the method of maximize the renewableenergy penetration with the integration of electric vehicles. Lopes et al. [23]investigated the dynamic behaviour of an isolated distribution grid when windpower and electric vehicles are presented. The objective is to quantify theamount of intermittent renewable energy resources that can be safely integratedinto the electric power system with the utilization of EVs’ storage capacity. Lundand Kempton [24] investigated the impact of using V2G technology to integratethe sustainable energy system. Two national energy systems are modelled; onefor Denmark including combined heat and power (CHP), the other is a similarlysized country without CHP. The model (EnergyPLAN) integrates energy forelectricity, transport and heat, includes hourly fluctuations in human needs andthe environment (wind resource and weather-driven need for heat). The resultsindicated that adding electric vehicles and V2G to these national energy systemsallows integration of much higher levels of wind electricity without excess electricproduction, and also greatly reduces national CO2 emissions.

In above discussion, mostly, these services can be practical in place only providedby a large fleet of electric vehicles. Fleet operator is widely proposed to aggregatethe large penetration of electric vehicles in the near future. Alternatively namesfor an electric vehicle fleet operator are used such as EV virtual power plant,EV aggregator, EV charging service provider or EV service provider. Fleet

2.1 Management of electric vehicle fleet and their business opportunities15

operators could be independent or integrated in an existing business functionof the energy supplier or distribution system operator. For more discussion interm of the fleet operator’s roles, the relationships between fleet operators andother actors in a smart grid context, and the communication standards used bythe fleet operator and electric vehicles, please refers to paper A.1.

To aggregate and attract the participation of electric vehicles, one incentive forEV owner is to minimize the charging cost, this could be facilitated by opti-mally charge the electric vehicles when the electricity price is lower. Withinthis scope, most the work [8,25,26] assume that the FO manages the electricitymarket participation of an EV fleet and presents a framework for optimal charg-ing or discharging of the electric vehicles. In addition, the electricity price ofthe day-ahead spot market and the regulation market and the driving patternsof the EV fleet are usually assumed to be known by the FO who is assumed tobe the price-taker in the electricity market in studies. By implementing linearprogramming or dynamic programming based cost minimization calculation, anoptimal charging schedule for electric vehicles will be generated by the fleetoperator. However, Kristoffersen et al. [27] also investigated the possibilitiesof EV management where the FO has a significant market share and can af-fect electricity prices by changing the load through charging and discharging.Besides studying the optimal charging from an EV fleet perspective, researchin [20, 28] showed how dynamic programming can be utilized by the individualEV controller to make an optimal charging schedule taking into account theelectricity market price. In [29], an intelligent charging method is also proposedfor individual electric vehicle which responds to time of use price and minimizethe charging cost.

In general, these studies focus on manage the electric vehicle centrally. In con-trast, although the decentralized control is a relative new application to EVfleet control, there are still a lot of efforts that have been done considering theamount of the articles. One notable study is about the valley filling study whichcan be seen in [30–33] where market based control strategies are used. In fact,the valley filling charging can date back to 1994 [34] where Ford argues thatvalley filling would allow the utility to meet anticipated additional loads forEV charging without additions to their existing resource plan. Therefore, heargued, the utility would experience an increase in profits from increased salesand better utilization of their generation equipment. In [30–33], the concepts aresimilar and the authors assumed that the EVs determines the charging patternindividually and are cost minimization.

In addition to the market based two way communication, by using one way pricesignal, it means that the EV controller do not need to propose and submit theircharging profile to the fleet operator, instead the fleet operator will anticipatetheir response to the dynamic price. The dynamic price ranges from simple

16 State of the art

time-of-use electricity rate [35, 36] to more varying hourly prices [37, 38]. Bothstudies [35,36] suggested that the time of use rates can be properly designed toreduce the peak demand as EVs penetrate the vehicle market. However, it isalso noted in [36] that the extent to which properly designed rates could assist inmaintaining grid reliability will remain open until empirically tested EV owner’sprice responsiveness through experiment pilots are known. Both studies [37,38]investigated the price elasticity of electricity consumers and these are also thekey issues in one-way price signal approach.

2.2 Control strategies for distribution grid con-gestion management

It is assumed that the distribution network has the capacity to allocate new loadswhen achieving the objectives discussed above. However, a large penetration ofelectric vehicles will bring some challenges to the utilities. The challenges usuallyinclude peak power issue, grid congestion, power losses, voltage drop et al. Muchresearch has been performed to study the intelligent EV load control and theireffect on the grid, which can be dated back to the early 1980s [39]. Heydt [39]argued that load management should be deployed to alleviate peak loading,which is measured in term of load factor improvement. In 1993, Rahman [40] andShrestha indicated that even low penetration levels of electric vehicles can createnew peak loads exceeding the natural peak if sufficient attention is not paid todistribute the charging load throughout the off-peak period. A penetration levelof 20% is found to be the upper limit which could be managed by distributingthe charging load. Basically, those studies mainly investigated the impacts byadding the new EV loading profile to the already existing load profile and seeingthe overall effect and then proposed the load shifting strategy. Recently, moretechnical parameters such as power losses, power qualities have been used forgrid congestion impact studies. In [7, 8, 41, 42], the impacts of electric vehicleson distribution networks are studied, the conclusions are that without chargingcoordination, the power consumption on a local scale can lead to grid problems.While the coordination of the charging can prevent the grid congestion, reducethe power losses, improve power quality etc.

In order to eliminate the grid congestions, three overall approaches have beenidentified in [43] to handle the integration of distribution grid congestion andelectric energy balance. Three listed approaches include integrated process,stepwise process form first electric energy balance, then grid congestion, andstepwise process form first grid congestion, then electric energy balance. Advan-tages and disadvantages of the three listed options are discussed in report [43].Different types of approaches including payment for the right to use capacity,

2.2 Control strategies for distribution grid congestion management 17

variable tariffs, direct control, a bid system etc., are discussed in the report.In the following, some articles are presented which can elaborate the discussionin [43] starting with the direct control, then with the market based control andfinally the price control method.

In study [44], the model being tested by Ericsson and partners is shown in Fig.2.3. With this proposal, The charge schedule can thus be selected directly bythe electricity utility to meet the owner’s or driver’s requirements, as well asspreading the demand over the course of the night. Further discussion are madeon the enabling technologies in the study.

1. Charging adapter is connected to dumb outlet

2. Car state changes to ready to charge

Public mobile network

4. Charging schedule request from the car to EVCS

5. EVCS submits charge schedule to utility

6. Utility validates car and provides detailed charge

schedule

3. Driver selects charging requirements

EVCS

Grid

7. EVCS sends detailed charging schedule to the car

8. Charging logic in the car executes charging schedule

Figure 2.3: The electric vehicle charging system, adapted from [44]

EVDSO RETCSP

Trip information

EV status

Trip

predictiion

Load curve boundaries

Preferred load curve

Load curve

optimization

Tim

e

EV charging schedules

Congestion cleared

load curve

Charging schedule

optimizer

Preferred load curves

Grid constraints

Preferred load curves

Grid constraints

Congestion cleared

Grid congestion

avoidance

Figure 2.4: The flow of information between the relevant actors, adapted from [8]

In [8], a complex scheduling problem involving the EV owners, the FO (charging

18 State of the art

service provider (CSP) in the article) and the DSO is analyzed. The approachrequires a complex interaction between the DSO and the CSP, on each interac-tion, the CSP will get a specific grid constraint from the DSO and add it intothe EV charging cost minimization problem. The results show that both theCSP and the EV owners can achieve the objectives of minimizing charging costsand fulfilling driving requirements without violating the grid constraints. Fig.2.4 illustrates the information flow between the relevant actors.

Lopes et al. [45] proposed a conceptual framework consisting of both a technicalgrid operation strategy and a market environment to integrate electric vehiclesinto the distribution systems, as shown in Fig. 2.5. The fleet operator is pro-posed to manage the electric vehicles and the fleet operators will prepare thebuy/sell bids into the electricity market. Having this defined, a prior interactionwith the DSO must exist to prevent the occurrence of congestion and voltageproblem in the distribution network. The smart charging algorithm is mainlydesigned for the operation of the DSO which can maximize the density of theEV deployment into the grid.

Technical Operation Market Operation

Generation System

Transmission System

Dis

trib

ution S

yste

m

Control

Level 1

Control

Level 2

Control

Level 3

DMS

CAMC

MGCC

CVC VC

GENCO

TSO

DSO

CAU

MGAU

Electricity

Retailer

EV

Owner

Parking

Facilities

Battery

Suppliers

Electricity

Consumer

Ele

ctr

icity M

ark

et O

pe

rato

rs

Players

Electric Energy

Technical Validation of the Market Negotiation for the

transmission system

Reserves

Reserves

Reserves

Electric Energy

Electric Energy

Electric

Energy

ParkingBattery

Replacement

Power systems

Fleet

operator

At the level ofControl (in normal system operation)

Control (in abnormal system operation) Comunicates with Buy offer

Sell offer Technical validation of the

market results

Figure 2.5: Technical management and market operation framework for EVintegration into electric power systems, adapted from [45]

Alternatively, several ways of solving the congestion problem have been sug-gested from market perspective. In study [46], Ipakchi pointed out that a tran-sition from a traditional view of power systems operation towards smart gridoperating paradigm is needed. It is also pointed that the higher penetrationof distributed resources will require a greater attention to distribution conges-


Wholesale

Market

Wholesale

Scheduling

Demand Side

Management

Supply

Management

(LSE)

Distribution

Operation

(UDC)

Generation

Transmissio

n

Transmissio

n

Substation

Sub

Transmissio

n

Distribution

Substation

Distribution7.8 kWh in 2 hours

uTag

Figure 2.6: Concept of micro-transactions for management of charging loads ofplug in electric vehicles, adapted from [46]

tion issues and the need for improved distribution automation and distributionmanagement capabilities. A transactive based approach is proposed to solvethe problems, Fig. 2.6 illustrates the concept. In the concept, A plug-in electricvehicle requests using 7.8 kWh of charging energy over the next two hours. Thisrequest can be presented as a demand transaction and sent to a Demand-SideManagement application operated by the utility distribution company. Knowingthe service delivery point to which the car charger is connected to, this applica-tion will check the available capacity of the secondary distribution transformer,lateral and feeder circuits and determine if this additional load will not impactthe circuit reliability and cause any adverse phase imbalances. The Demand-Side Management application will then schedule the charging for the requestedtime period. At the same time, the application may receive many more informa-tion such charging requests that have to be checked, and in aggregate they haveto be coordinated with wholesale scheduling at substation supplying the feedersto ensure adequate supply. Each of these actions could be modeled as a transac-tion between a consumer system, a utility decision support system, distributionfield equipment and supply scheduling system in an aggregate form.

Transactive control is a kind of market based control technology and consideredas a specific instantiation of a GridWise Architecture Council [47,48]. The intent

20 State of the art

of transactive control is to standardize a scalable, distributed mechanism for ex-changing information about generation, loads, constraints and responsive assetsover dynamic, real-time forecasting periods using economic incentive signaling.Fig. 2.7 from [49] lists the key issues of the concept. It presented an exam-ple that precise, stable control of congested grid nodes can be derived from (1)customer price-responsive controls that (2) express their available flexibility to(3) a price discovery mechanism. In other words, the GridWise initiative haveadopted an agent based computational economics modeling for incorporatingthe market mechanisms that allow the system to evolve over time in responseto market dynamics.

1. Price-response device control

2. Customer price-flexibility curve

Load (kW)

Price ($/kWh )

Price ($/kWh )

Price ($/kWh )

Load (kW)

Indoor temperature

Price

Bid

Tair Tset

Max load

Base load

Charge battery

Water heater

AC

Discharge battery

3. Price-discovery mechanism

Pclear

Pwholesale

Rated node capacity

Node supply curve

Demand curve

Figure 2.7: Key issues in a transactive control, adapted from [49]

PowerMatcher 1 is another good example of a market-based control conceptfor supply and demand matching in electricity networks. It is discussed in[50] that the PowerMatcher technology is based on multi-agent systems andelectronic markets which form an appropriate technology needed for controland coordination tasks in the future electricity network. Background theoriesused are control theory and micro-economics, unified in market-based controltheory and it is presented in [51], one of the conclusions from the study is thatcomputational economies with dynamic pricing mechanisms are able to handlescarce resources for control adaptively in ways that are optimal locally as wellas globally. The basic structures and agents are illustrated in Fig. 2.8, for theroles and functions of the agents, please refer to PowerMatcher website.

In addition to the proposed market based approach, price control is also exten-sively discussed in several studies [9,43,52]. Several tariff signals including time

1http://www.powermatcher.net/


Auctioneer

Agent

Concentrator

Agent

Price

Bid

Concentrator

Agent

Bid

Price

Bid

PriceDevice

Price

Bid

Bid

Price

Price

Bid

Figure 2.8: Concept of PowerMatcher technology, adapted from [50]

of use tariffs, critical peak pricing, variable tariff and dynamic tariff have beengeneral discussed and compared in [43,52]. The comparison are discussed fromthe activation, timing of price determination, tariff characteristics, price varia-tions, advantages and disadvantages perspectives. In both reports, a bid-less,day-ahead setup market model is discussed, Fig. 2.9 illustrates the model. Themodel has been tested in study [9] and the results indicated that it could beuseful to alleviate the grid congestion.

Balance responsbile

Local grid companyCustomer

Step 1: 10:00Grid company sends an

indication of the forecasted state of the distribution grid for the next 24 hour period

Step2: 12:00Balance responsible bids in

on the spot market. Adjusted for local

congestion if significant

Step3: 13:00Balance responsible is

informed of accepted spot price bids

Step4: 14:00VPP: Balance responsible

controls heat pumps based on accepted bids and customer parameters

Figure 2.9: A bid-less day-ahead setup market model, adapted from [52]

22 State of the art

2.3 Multi-agents systems for smart distributiongrid operation and simulations

Multi-agent systems (MAS) have been widely proposed for several studies in thecontext of power systems, such as power system restoration [53], power systemoperation and control [54], and electricity markets modeling and analyzing [55,56]. More recently, the multi-agent concept is proposed for distribution systemoperation and control [57–61], especially, considering the capacity managementwith a large penetration of electric vehicles [62,63] and the capacity managementwith more general loads [64].

The authors [62] proposed a distributed, multi-agent electric vehicle chargingcontrol method based on Nash certainty equivalence principle that considers dis-tribution network impacts. Four types of agents are included in the study, EVaggregator agent, regional aggregation agent, microgrid aggregation agent andcluster of vehicles controller agent, and vehicle controller agents. In the non-cooperative, dynamic game, all the vehicles controller agents decide the strategythat minimizes his own objective functions. The up-level agents coordinate ve-hicles controller agents’ charging behaviour by changing the price signal. Theprice signal is a reflection of congestion conditions. The results indicate thatthe proposed approach allocates electric vehicle energy requirements efficientlyduring off-peak hours which achieves effectively valley filling and also leads tomaximization of load factor and minimization of energy losses. The authorsin [63] used the MAS to design a distributed, modular, coordinated and collab-orative intelligent charging network with the objective of pro-actively schedulingthe charging of up to fifty electric vehicles as well as eliminating the grid over-loading issue. The study mainly considered how the electricity is distributed tothe multiple charging point agent under one local power manager agent and thisis done by an auction mechanism. Each charging point agent makes a bid forthe energy in the next 15 minutes until it get the desired state of charge of thebattery, then the local power manager agent sorts out the orders to determinewhich electric vehicle can be charged during the time slot. In [64], an active dis-tribution network (ADN) is presented with its actors and their objectives. Themulti-agent technology is proposed for the normal operation of the ADN, inwhich the auctioneer agent (placed at the MV/LV transformer) communicateswith the device agent by sending the price signal and receiving the bid curve.Further on, capacity management is investigated by transforming the bid curvesof the device agents.

In general, the above studies [62–64] adopted an agent-based computationaleconomics modeling 2 for incorporating market mechanisms to allocate the grid

2Please see the discussion regarding GridWise and PowerMatcher concept review.

2.4 Summary 23

capacities and the results indicated the agent-based market mechanisms solvethe grid congestion efficiently. However, in term of simulations, much workneeds to be studied, as discussed in [57], currently available software tools pro-vides either modeling of dynamic electric power networks or implementationof intelligent system techniques such as intelligent agents. Therefore, a soft-ware platform which can simulate multi-agents based control strategies thatincorporates power system dynamics and intelligent control strategies is highlyrequired. In [57], the author developed a software platform based on the inte-gration of JADE and DigSilent PowerFactory which can enables design, test,and verification of agents based control technologies. Besides, in [65], a simula-tion framework named mosaik is developed to facilitate the simulations runningseveral separately software together, the purpose of the developed framework isto reuse existing models in a common context to simulate complex smart Gridscenarios in order to evaluate control strategies, such as agent based controlstrategies.

2.4 Summary

The overall state of the art of the technology for capturing the business oppor-tunities provided by electric vehicle fleet is quite feasible and promising. But asa commercial actor, which control strategies is optimal, and what is the implica-tions for choosing one control strategies to manage the electric vehicle fleet? Atthe same time, the commercial actor needs to coordinate with the distributionsystem operator to prevent the congestion before bidding into the conventionalmarkets, several approaches have been reviewed and there is a need to devisethe suitable one for this context. In addition to this, to simulate and evaluatethe control strategies developed for distribution grid congestion managementwith EV integration, whether multi-agents based technology is a good approachand if yes, how to develop a multi-agent systems? The need to answer suchquestions have motivated the choice of research topics for this thesis.

Chapter 3

Optimization and control forintegrating electric vehicles

In this chapter the main results concerning the first research topic are summa-rized. The main results and details have been published in separate papers A.1,A.2, A.3, and A.4 that are included in appendices of this report. The chap-ter starts with some background knowledge including the electricity market,battery modeling, and vehicle driving pattern, because these are the primaryinputs that are needed in an electric vehicle optimal charging problem. Then, wepresent the main results including a comprehensive review on the three controlstrategies developed for electric vehicle integration and four developed meth-ods during the PhD study, i.e., problem formulations of electric vehicle optimalcharging schedule by linear programming, dynamic programming, mix integerlinear programming and statistic modeling based techniques.

3.1 Background knowledge

3.1.1 Nordic electricity market

In the Danish Edison project 1, an overall discussion is given on the currentNordic electricity market and how the electric vehicles can be integrated intothe current and future markets [43,66]. The reader is referred to the two reportsfor more details. Here we briefly introduce the spot market and the regulationmarket since they are most relevant to the thesis.

1http://www.edison-net.dk/

26 Optimization and control for integrating electric vehicles

3.1.1.1 Spot market

In the Nordic electric power market, energy is exchanged by direct tradingamongst player (bilateral trade) and via the Nordic power exchange, NordPool.There are two main markets for energy exchange within NordPool that areElspot for day-ahead trading and Elbas for the balancing market. Elspot isa day-ahead market where hourly exchanges are traded. The way of calculat-ing the price is called double auctions, as both the buyers and the sellers havesubmitted bids. At noon, NordPool Spot’s computer in Oslo starts calculatingthe day-ahead price. Having finished the calculation, NordPool Spot publishesthe prices. At the same time, NordPool Spots reports to the participants howmuch electricity they have bought or sold for each hour of the following day.These reports on buying and selling are also sent to the TSO in the NordPoolspot area. The TSO use this information, when they later calculate the balanc-ing energy for each participant. Apart from calculating day-ahead prices, theElspot market is also used to carry out day-ahead congestion management in theNordic region. The day-ahead congestion management system is called marketsplitting. More discussion regarding market splitting can be seen in [67,68].

3.1.1.2 Regulating power market

The regulating power market is managed by the transmission system operator2 in order to obtain ancillary services in the transmission grid. It may happenthat the consumption exceeds (or lags behind) the generation. In this case, thefrequency of the alternating current will fall to (exceed) a value below (above)50 Hz. As renewable energy will become an increasing important resource forreducing emissions from fossil fuels, production will become more intermittent,and therefore it is anticipated that the need for regulating power will increase[43,66].

3.1.2 Battery modeling

Basically, there are two ways to model the charging characteristics of electricvehicles, i.e., the battery. One is the individual battery pack model, anotheris the aggregated or cell based model. For simplicity, most of the studies re-viewed considered it as a battery pack when investigating the optimal chargingor discharging problem. Currently, most battery model studies focus on threedifferent characteristics [69,70]:

2http://www.energinet.dk/EN/El/Systemydelser-for-el/Sider/Systemydelserforel.aspx

3.1 Background knowledge 27

• The first and most commonly used model is termed as a performance or acharge model and focuses on modelling the state of charge of the battery,which is the single most important parameter in system assessments.

• The second type of model is the voltage model, which is used to modelthe terminal voltage so that it can be used in more detailed modelling ofthe battery management system and the more detailed calculation of thelosses in the battery.

• The third type of model is the lifetime model used for assessing the impactof a particular operating scheme on the expected lifetime of the battery.

The present study mainly focuses on smart charging of electric vehicle andtherefore we present more introduction on the first characteristic, i.e., modelingthe state of charge of a battery during the operation.

Rint

=

I2

U2Uoc

Figure 3.1: Equivalent circuit of a battery

A basic physical model of a battery can be derived by considering an equivalentcircuit of the system like the one shown in Fig. 3.1. The steady state batteryequivalent circuit has been applied mainly for various lead acid batteries, butalso for nickel cadmium, nickel metal hydride and lithium-ion batteries. Inthis circuit, the battery is represented by an voltage source in series with aninternal resistance. Kirchhoff’s law for the equivalent circuit yields the followingequations:

Uoc(t)−Rint(t).I2(t) = U2(t). (3.1)


Both the voltage source Uoc(t) and the internal resistance Rint(t) are dependenton the state of charge (soc) 3 of the battery.

Normally, two ways are used to characterize the capacity of a battery, kWh andAh 4. If the dynamics of the state of charge of the battery is calculated byadding kWh into the available capacity, it usually starts with the calculationof the internal power of the battery which can be approximated linearly ornonlinearly 5 with the external charging power, like the one presented in thestudy [8,25], also described in the section B.1. The studies [8,25] have shown thatthe difference between the two charging schedules using linear approximationand nonlinear approximation (second-order Taylor series expansion) is minorand indicates that the linear approximation is sufficient and the benefit of usinga nonlinear approximation does not justify the increase in computation time.If the dynamic of the state of charge of the battery is characterized by thecalculation of the dynamics of the electric charge, like the one used in study[20,28], also described in the section B.1, dynamic programming is usually usedto formulate the optimal charging of electric vehicle.

3.1.3 Driving pattern and associated electricity consump-tion

The analysis of driving pattern can be divided into two main directions:

• Utilization of electric vehicles, in other words, when and how long theelectric vehicle will be used in the next scheduling period, e.g., 24 hours inthe present study. This is because when and how long decide the energythat need to be procured or charged for the next scheduling period.

• Location of electric vehicles when charging and how many of them will becharged at a time, because the location of the electric vehicles inside thenetwork will determine where the grid will be possibly congested.

In most studies [8,20,25], the authors assume that the fleet operator knows theusers’ driving patterns and thereby can forecast the electricity demand. Thereare few studies on investigating the driving pattern. Kristoffersen et al. [27]

3Usually, the abbreviation for state of charge should be capital letter, i.e., SOC. Here, weuse ’soc’ because it will be used as a variable in section B.1.

4kWh=Ah*voltage5Using linear approximation, the internal power is assumed to be equal to the external

power; using nonlinear approximation, the internal losses in the battery can not be neglected,therefore, the internal power is not equal to the external power

3.2 Main results 29

investigated a method to construct driving patterns from the historic data inDanish case. By clustering the survey data on the vehicle fleet in Western Den-mark (January 2006-December 2007), representative driving patterns for eachvehicle user are constructed. S. Shahidinejad et al. [71] developed a daily dutycycle which provides a complete data set for optimization of energy require-ments of users. And furthermore, this information is used to analyse the impactof daytime charging by a fleet of plug-in electric vehicles on the electric utilitygrid that may create a peak demand during the day to be met by the local utilitygrid. Normally, intra city or short term driving patterns are largely predictabledue to fixed working hours and fixed business schedules and routes.

In the present thesis, the driving pattern is based on the 2003AKTA Survey[72], where 360 cars in Copenhagen were tracked using GPS from 14 to 100days. Each data file includes starting and finishing time, and the correspondingduration and distance. The original data is transferred into 15 minutes intervaldriving energy requirements based on the assumption of that an electric vehiclewill use 15 kWh per 100 km (typically, the number ranges from 11 kWh to18 kWh per 100 km). Some artificial driving data of the electric vehicles havebeen generated from the database based on some facts observed in study [73].In [73], a Danish driving pattern analysis is presented which listed that theaverage driving distance in Denmark is 42.7 km per day. With the assumptionof 0.15 kWh/km for energy used per km of electric vehicles, one can deducethat the monthly energy requirement for an electric vehicle will be around 192kWh (42.7 km*30*0.15 kWh/km). Using Nissan Leaf (EV battery capacity is24 kWh) as an example, this would imply that the users need to charge the Leafaround 8 times (192 kWh/24 kWh). However, owners will rarely fully dischargetheir Leaf before recharging it. For discussion purposes, it is assumed users willcharge the electric vehicle 20 times per month which implies that each time 9.6kWh energy will be procured. In the current case, they would imply aroundfive hours charging (9.6 kWh/2.3 kW). Using 9.6 kWh as an average number,the driving data is randomly chosen and transferred into a 15 minutes intervaldriving energy requirements.

3.2 Main results

3.2.1 Three control strategies for integrating electric ve-hicles

As discussed in paper A.1, also reviewed in section 2.1, several business oppor-tunities provided by EV fleet have been identified such as providing ancillary


services to transmission system operation and storing service to renewable en-ergy producers. To capture the business opportunities, fleet operator has beenproposed to manage EVs. Fleet operator [74, 75] could be independent or in-tegrated in an existing business function of the energy supplier. Furthermore,Fleet operator needs to coordinate with the distribution system operator tomanage the congestions in the distribution network. In principle, two types ofcontrol architectures can be used by FOs when aiming at realizing the businessopportunities, named centralized and decentralized control. In both control ar-chitectures, the grid constraint from the distribution grid should be considered.

FO operation

Modeling the state of charging of the battery with linear or nonlinear approximation

life time assessing model, battery degration analysis

Day-ahead market Regulation power market Other reserve market

Forecasting driving requirement of EV owner

Location prediction of Evs Normal vehicle’s historical

data are used Normal distribution,

Monte-Carlo based method

Thermal constraint of cable and transformers

Voltage constraint

Battery Model

EV FO

Electricity Market

Driving Pattern

Grid Constraints

EV1 EV2 EVNEVi

Control Signal

Status information

Information source

Decision making

Figure 3.2: Centralized control: primary inputs and output of the electric vehiclefleet operator

In centralized control, electric vehicles are aggregated and controlled by fleetoperator directly. e.g., by dictating the charging schedules. In decentralizedarchitecture, the control is implemented in the form of price signal, i.e. theindividual EV optimizes the charging schedule based on the electricity priceinformation made available to them either from FO or the utility. Figure 3.2depicts the mainly four inputs when making the control strategies in the caseof centralized control. The FO obtains all the relevant information includingthe battery model, the driving patterns, the grid constraints and the electricityprice and centrally makes the charging schedule for each EV. In contrast, in thecontext of decentralized charging, the FO uses price signal to coordinate the EVusers’ charging behavior. Two methods of implementing decentralized chargingcontrol are summarized. The scheme of the information flow in decentralized

3.2 Main results 31

charging control is presented in Fig 3.3. In Fig. 3.3 (a), the basic idea for marketbased control strategy is that EVs update their charging profiles independentlygiven the price signal; the FO guides their updates by altering the price signal.Several iterations are usually required for the implementation. In Fig. 3.3 (b),the price control method requires FO to predict the users’ response to the prices.The price signal can be designed simply as time-of-use price or as dynamic prices.

Charging profile

of vehicle 1,...N

Price

information

EV FO perspective

Send the updated price/

aggregated schedule

EV owner perspective

Intelligent Individual controller

Sophisticated control equipment

Send back the charging profile

EV FO

EV1 EVNEVi

Price

information

EV FO

EV1 EVNEVi

EV FO perspective

Predict the user’s response/price

elasticity

Determine the optimal price signal

EV owner perspective

Programmable appliance timers

Price flexible consumer

A more sophisticate technology for

price response

Price control strategy Market based control strategy

(a) (b)

Figure 3.3: Decentralized control: a schematic view of the information flowbelow the fleet operator and the electric vehicles

3.2.2 Formulation of the optimal charging of electric ve-hicles by centralized control architecture

In the following, we introduce the method to formulate the optimal chargingschedule of electric vehicles based on linear programming, dynamical program-ming, and mixed integer programming techniques.


3.2.2.1 Linear programming based optimal charging schedule gener-ation

As explained in section 3.1.2, this study uses the linear approximation method.Linear programming approach is utilized to optimize the charging schedule of anEV fleet both taking into account spot price and individual EV driving require-ment with the goal of minimizing charging costs. A simplified version of Fig. 3.2is used to guide the control algorithm design. The charging schedules of the EVfleet are optimized individually and then each EV’ schedule is summed, since itis considered that the individual’s driving requirements should be fulfilled andcalculated separately. The formulation is shown in the following:

minimize

NT∑

i=1

Φj,i · Pj,i · t, j = 1, ..., NEk

subject to

SOC0,j +nT∑i=1

Pj,i · tj,i ≥ SOCMin,j +nT−1∑i=0

Ed,i+1

SOC0,j +nT∑i=1

Pj,i · tj,i ≤ w · Ecap,j +nT+1∑i=2

Ed,i−1

0 ≤ Pj,i · tj,i ≤ Pmax,j · tj,i, i = 1, ..., NT

(3.2)

With the above optimization problem, the FO can generate a unique energyschedule for EV owner; the sum of the individual EV energy schedule will bedenoted as PEk,i, and

PEk,i =

NEk∑

j=1

Pj,i, k = 1, ..., NB , i = 1, ..., NT ,

where

NEk Number of EVs under FO k.

NT Number of time slot in the scheduling period.

NB Number of FOs.

j Index for the number of EVs under each FO, j = 1, 2, ..., NEk .

i Index of time slot in the scheduling period, i = 1, 2, ..., nT , ..., NT .

k Index for the number of FOs, k = 1, ..., NB .

3.2 Main results 33

Φj,i Predicted day-ahead electricity market price vector.

Pj,i Decision variable vector.

t Length of each time slot.

PEk,i Power requirements of EVs of each FO in each time slot.

SOC0,j Initial state of charge of individual EV.

SOCMin,j Recommended minimum state of charge of the EV.

Ed The predicted individual EV owner’s driving requirement.

Pmax,j Charge rate in term of energy of individual EV.

w ∗ Ecap,j Recommended maximum state of charge of the EV, where w is theparameter which express the charging behavior of the battery of the EVis a linear process, Ecap,j is the capacity of the battery of the EV.

In Eq. (3.2), the first constraint means that the available energy in the batteryshould be greater than or equal to the energy requirement for the next trip.The second constraint indicates that the available energy in the battery shouldbe less than or equal to the power capacity of the battery. The third constraintrepresents that the charging rate is less than or equal to its maximum powerrate of a charger. The physical meaning of the decision variable vector Pj,i is tomake a decision to distribute/charge the power on the certain time slots, wherethe charging cost can be minimized. The solutions of some numerical studiesare presented in paper A.2, A.6.

3.2.2.2 Dynamic programming based optimal charging schedule gen-eration

As explained in section 3.1.2, the study presented in paper A.3 considers thedynamics of the state of charge of the battery using electric charge. The controlarchitecture used in the study is also a simplified version of the one shown in Fig3.2. For the day-ahead scheduling, the time horizon [0, N ] of a day is discretizedinto equidistant time intervals [k, k+1] with k = 0, . . . , N−1. It is assumed thatthe time interval is ∆t. This problem is addressed by considering the followingdiscrete system which describes the battery:

xk+1 = T (xk, uk, k) (3.3)


State variable xk represents the state of charge of the battery at time k. xkis not only discrete in time (index k) but also in value. Any value has to beincluded in the predefined set X, which can be calculated by a function of chargeQk and total capacity Qmax.

xk =QkQmax

(3.4)

uk in equation 3.3 is the control variable, which is dimensionless and discrete.uk is multiplied with the maximum available charge power (Pmax) when electricvehicle is connected with the grid. The values of uk are fixed at 0 when driving,while these values range from 0 to 1 when electric vehicle is connected to thegrid. If Uplug is the set that covers all possible values of uk, its discretizationmay be described as follows:

uk =

{uk ∈ Uplug, k ∈ Kplug

uk = 0, k ∈ Kdriv(3.5)

Kplug is a set of indices k within the time periods where the vehicle is pluggedin, while Kdriv refers to the driving intervals. The summation of the numberof elements in Kplug and Kdriv is N, the total number of time intervals. Anyindex k in Kplug or Kdriv has to be element of the predefined set K.

k ∈ K = {Kplug,Kdriv} (3.6)

A specific control strategy is represented by

u = {u0, u1, u2, ..., uN−1} (3.7)

Any value of uk has to be element of a predefined set U, which is known as theset of admissible decision. The total cost of a sequence, fU0 , is given by the costof the final step, fN (xN ), plus the cost for all other previous steps, vk(xk, uk, k),then we have:

fU0 (x0) = fN (xN ) + ΣN−1k=1 vk(xk, uk, k) (3.8)

The optimal control strategy u∗ = {u∗0, u∗1, u∗2, . . . , u∗N−1} minimizes the costfunction 3.8 and can be determined by the dynamic programming. For casestudies and their solutions, please refer to paper A.3.

3.2.2.3 Mixed integer linear programming based optimal chargingand discharging schedule making

In order to investigate the economics of vehicle to grid technology, e.g., providingancillary service to the regulating power market, four different charging schemes

3.2 Main results 35

including night charging, night charging with V2G, 24 hour charging and 24 hourcharging with V2G are studied in paper A.4 to compare their annual cost.

Two steps are used in the study. In the first step, the numerical comparison offour charging schemes only includes charging cost (charging cost is equal to costof electricity procured minus profits of implementing V2G). This is obtained bysolving a mixed integer programming problem with the purpose of minimizingthe charging cost by taking into account the users’ driving needs and the prac-tical limitations of the EV battery (capacity of the battery, recommended stateof charge range). A formulation of the problem is shown below:

min

N∑

i=1

{∆Ec(i) · Φ(i) · u1(i)

ηc+ ∆Ed(i) · Φ(i) · u2(i) · ηd

}

subject to

E(i) = E0 +∑k=ik=1{∆Ec(k) · u1(k) + ∆Ed(k) · u2(k)− Ed(k) · u3(k)}

δmin · Ecap ≤ E(i) ≤ δmax · EcapEd(i+ 1) · u3(i+ 1) ≥ E(i)

0 ≤ ∆Ec(i) ≤ Pc,max · ηc ·∆t−Pd,max

ηd≤ ∆Ed(i) ≤ 0

u1(i) + u2(i) + u3(i) = 1(3.9)

where Φ(i) is the electricity price and Ed(i) denotes driving energy requirements.The decision variables ∆Ec(i) and ∆Ed(i) represent the energy charged into anddischarged from the battery in each time interval respectively, while the otherthree binary variables u1(i), u2(i), u3(i) indicate the on/off status of charging,vehicle to grid (discharging), and driving for each corresponding time inter-val. To facilitate the formulation, an intermediate variable E(i) is introducedrepresenting the energy level of the battery at the end of each time interval.Parameters Ecap and E0 represent the nominal energy capacity and the initialenergy of the battery in the planning period, while the charging and discharg-ing efficiency are represented by ηc and ηd. The maximum power exchangedbetween the EV inverter and the electrical grid are expressed by Pc,max charg-ing and Pd,max discharging respectively, which constrains the maximum energyexchanged between the electric vehicle and the grid. Concerning battery life,δmin and δmax are further introduced to represent the manufacturer recom-mended sate of charge (SOC) range. Explanations of the inequality constraintscan be found in section 3.2.2.1. Based on the model, optimal charging planswith 5 minutes resolutions are derived.

In the second step, or the post processing stage, the rainflow counting algo-


rithm is implemented to assess the lifetime usage of a lithium-ion EV batteryfor the four charging schemes. By applying the rainflow counting algorithm tothe monthly charging profiles calculated in the first step, and take the relation-ship between the number of cycles and depth of discharge, the battery lifetimeconsumption for different charging schemes are calculated. (For the details ofrainflow counting algorithms, please refer to paper A.4.) Finally, a simple ap-proach is introduced to roughly estimate the annual cost for different chargingschemes:

Cann = (Ccapacity + Ccharging)/Lexp (3.10)

where Ccapacity and Ccharging represent the capital cost of the battery and charg-ing cost incurred during the battery lifetime respectively, and Lexp indicates theexpected lifetime for different charging schemes. This complete the calculationof the annual cost. The study illustrated that the night charging scheme is thecheapest solution among four different charging schemes. For parameters valuesand calculation results etc., they are presented in paper A.4.

3.2.2.4 Algorithm comparisons

Together with the observations from other research, the readers are referred topaper A.1, table 3 to find the comparison among the chosen algorithms suchas linear programming, dynamic programming, quadratic programming, andstochastic programming. The comparisons include the computation time, thecertainty of performance, and the applicability.

3.2.3 Formulation of the optimal charging of electric ve-hicles using decentralized control architecture

For the decentralized control architecture, price control is studied. As illustratedin Fig. 3.3 (b), the price control method requires FO to predict the users’response to the prices. In order to study how the price can regulate the EV user’scharging behavior, a statistical model of demand elasticity proposed in [38] isused for this study. In the model, the marginal utility function of the loads isrealized by the following parametric stochastic process:

r(t) =

{β − δ(t− α), α ≤ t ≤ α+ γ0, otherwise

(3.11)

where α, β, γ, δ are random variables that describes the different characteristicsof utility function as follows:

3.3 Summary and discussions 37

a) α stands for the time slot that a task is initially requested, which also reflectsthe task distribution.

b) β is the initial marginal utility, which stands for the magnitude of themarginal utility.

c) γ is the tolerable delay, which determines the maximum delay that a user cantolerate to finish a task.

d) δ means the utility decay rate, which represents the cost of inconvenience bythe delay.

Under this model, the scheduling of each individual task is now a random eventwhose probability distribution is controlled by the stochastic process r(t). Theaggregated demand curve can be estimated through expectation with respect tothe distribution of r(t). Note that some assumptions have been made before,such as the time period of the scheduling is divided into T time slots, the totalM individual tasks m : m = 1, ...,M of different electric vehicles that are tobe initialized by all the users within the scheduling period, and each task willconsumer xm kWh energy, all in all, X0 is the total energy consumption tobe scheduled in the target scheduling period. Furthermore, it is assumed thateach task can be completed within one time slot; therefore, tasks that haveduration longer than one time slot will be decomposed into multiple tasks thatare considered independently. A case study is performed and the simulationsare presented in section B.2.

3.3 Summary and discussions

This chapter mainly investigates the control strategies for commercial actors,such as fleet operators. However, some methods can also be used by the utilityto coordinate the charging profiles of electric vehicles. The control strategiesinvestigated including centralized control and price control. For centralizedcontrol, three algorithms have been used for the methods development. It isrecommended that linear programming based technique is considered for char-acterizing the optimal charging scheduling problems. The recommendation isbased its overall performance and its fast computational time. Regarding pricecontrol, it is demonstrated that price signal is an effective way of regulatingthe charging behavior of electric vehicles, however, more research is needed indesigning a proper rate which can mitigate the uncertainty caused by using theprice control.

38

Considering fleet operators’ practical operations, it is assumed that EVs needto subscribe to one FO, maybe in the form of signing a contract that is validfor certain time period. Such subscription would possibly following the existinggeographical areas, i.e., the neighborhood supplied by one FO, under one sub-station. The mobility of electric vehicle, in such context, will also require theroaming-related agreement/standards among different FOs as well as an stan-dardized information and communication infrastructure, in order to make surethe fleet operators can access the EV information immediately when the electricvehicles switch fleet operators.

In order to generate an optimal charging schedule, forecasted electricity priceand predicted EVs driving pattern are essential, fortunately, they are pre-dictable. Among the presented control strategies, it depends on the contextto choose the right method (suppose the electric vehicles can be either directcontrolled or be price coordinated). Some business scenarios are consideredhere:

• In scenario 1, FOs only participate in the day ahead market with theobjective of minimizing the energy cost, it is recommend that direct controlis most efficient and economical, together with the linear programmingbased technology.

• In scenario 2, the FO would like to participate in both the day-ahead spotmarket and the regulating power market, direct control is still our firstoption considering it high certainty in term of biding resources preparationand activation.

• In scenario 3, the FO will help the DSO to reduce the peak load, it issuggested that either direct control or price control can be used, eachmethod has its advantages and disadvantages, for example, price controlis easy to be implemented comparing to direct control while direct controlcan ensure a low risk.

Such illustrative scenarios are presented to show that although the options forfleet operators are not one and only, the difference does exist considering differ-ent questions/contexts.

Chapter 4

Market based control fordistribution grid congestion

management

In this chapter the main results concerning the second research topic are sum-marized. The main results have been published in separate papers A.5, A.6,A.7, A.8, included in appendices of this report. The chapter starts with theformulation of congestion problems in distribution network. Then, the chapterintroduces the background knowledge for market based control method which isthe solution for distribution grid congestions in the present thesis. Finally, thechapter presents the main results of the second research topic.

4.1 Formulation of the congestion problems indistribution network

In this section, first, the impact of electric vehicle’s charging as an extra loadon the distribution grids is illustrated. Then, congestions management in dis-tribution network operation and its enabling functions in a smart distributiongrid paradigm are presented.

40 Market based control for distribution grid congestion management

4.1.1 Electricity consumption by electric vehicles and itsimpact on the distribution grids

It is mainly regarded that electric vehicle as a kind of new load will mainly bringchallenges to the distribution network level such as congestions and voltagedrops, while some discussions are taken from the perspective that simultaneouscharging of electric vehicle at system peak could result in shortage or create aneed for large new investment in generating capacity. To explain these points,we provide some illustrative calculations based on the parameters or estimatednumbers related to Danish power systems [2,73,76,77]. Report series [2,76] pub-lished by the Danish authorities have estimated the number of electric vehiclecould be around 600,000 by 2025. Since there are around 70 distribution com-panies in Denmark, the average number of electric vehicle for each distributioncompanies would accordingly be around 8600 1. Two charging rate scenarios areconsidered in the distribution networks (residential area) which include 2.3 kW(AC 10 A*230 V) and 3.7 kW (AC 16 A*230 V). Considering the worst case,i.e., the users simultaneously charging the electric vehicle, this will introduceloads of demand around 20 MW and 31 MW correspondingly to the distribu-tion companies. Generally, the utilities have the capability to address this loadgrowth.

However, as discussed in [36, 52, 77], congestion might happen in the middle orlow voltage grid. Each distribution grid has a different history of development,such as in some cases congestion is first expected to emerge in the mediumvoltage grid, while in other grids the low voltage grid is considered to be morecritical. In [36], the authors presented a case in the United States where thetypical transformer serves anywhere from four to ten homes. In an area wherethe household load is 3 kilowatts (kW), the new load of an electric vehicle couldeasily double and become 6 kW per house. In areas where it is currently 6kW per house, it could rise by 50 percent and become 9 kW per house (ormore). In [77], a Danish case is presented to illustrate the changes of the loadprofile with one hundred percent penetration of EV and it is clearly shown thatwithout any control, the transformer and the cable in a low voltage grid will beoperated under pressure. As discussed in section 3.1.3, an electric vehicle ownerin Denmark will probably charge the electric vehicle 20 times per month. Eachtime, the electric vehicle need to be charged five hours. Without any control,the distribution systems are operated with high load profile nearly fifty percentof a year.

Basically, the above discussion is an illustrative calculation on the load profiles of

1It is noticed that Dong Energy Distribution is the largest distribution companies in Den-mark, certainly the share of the number of electric vehicle will be high, this implies that otherrelatively smaller distribution companies share less comparing to the average numbers.

4.1 Formulation of the congestion problems in distribution network 41

the distribution system. Some complicated models [78–81] have been proposedto address the load demand in distribution system considering the electric vehiclecharging behavior and can give more accurate estimation on the impact of theEV loads to the distribution network.

4.1.2 Congestions management in distribution network op-eration and its enabling functions

Typically, the challenges in the distribution grid caused by demand spikes aresolved by expanding the grid to fit the size and the patterns of demand. As analternative, inspired by the congestion management at the transmission level 2,described in section B.3. This study will consider allocation of the capacity ofthe distribution network according to economic principles. In order to identifyand solve congestion problems caused by the increasing new loads such as elec-tric vehicles, heat pumps, the distribution system operator requires additionalinformation about the current and anticipated operating state of the distribu-tion grids. This implies a need for measurement equipment and/or technologyenabling identification and anticipation of load patterns and grid ’bottlenecks’.Key operations for DSO congestion management in a smart grid operation wouldbe:

• Demand forecasting (conventional loads and new demands such as EV andheat pump)

• Generation forecasting

• Grid state estimation

• Online grid measurements

• Real-time intervention in case of unexpected deviations challenging gridreliability

• Meter data collection and analysis

However, DSOs’ tasks in conventional system operation [82], are mostly focusedon ‘off-line’ tasks related to asset management and maintenance during normalsystem conditions. The primary objective under emergency conditions is toorganize restoration of the network as quickly as possible. Distribution systems

2The methods for congestion management in the transmission system is described in sectionB.3. In addition, the similarities and differences of congestion management in the transmissionsystem and distribution networks are also compared in that section.


today tend to be weakly monitored as compared to transmission grids, and theyare controlled in a decentralized fashion on the basis of pre-configured localcontrols (e.g. by means of grid codes and protection settings).

To illustrate a future operation scenario or smart grid operation scenario witha higher level of automation, in [83, 84], smart grid for distribution systemshas been envisioned and the benefits and challenges of implementing the manydifferent distribution automation functions are also presented. It is believed thatthe functions required in a smart grid operation can be extended with additionalonline and data intensive acquisition from the functions existing in the currentdistribution grid operations.

4.2 Background knowledge of the market basedcontrol solution

4.2.1 What is market based control?

Market based control [51,85–87], is a paradigm for controlling complex systemswith conflicting resources. It typically includes the features found in a marketsuch as decentralized decision making and interacting agents. As discussedin [85], the “control” in a market based system emerges from the individualgoals of the agents rather than having a goal imposed from above. Comparedto centralized control, there is no need for any of the agents in a market basedsystem to know all the parameters of the system in order for the overall systemto function smoothly. Instead, in the market, with very little information, i.e.,price, it is possible to facilitate allocation of resource. In fact, the market-basedapproach has been recommended to be used in the power distribution system,such as the discussion in joint research center European Forum 3 or in theresearch [88].

There are many mechanisms which can be used to find the clearing prices suchas the general equilibrium market mechanism explained in the following section,the auction based approach and contract net protocols described in section B.4which are not the focus in the present thesis. The general equilibrium marketmechanism is mainly considered due to the following reasons:

• Considering the size of the market in the distribution system level 4, it is

3European commission Joint research centre, Scientific support to capacity markets andthe integration of renewables, Brussels (BE) - 22/07/13

4Imagine that the highly possible actors in this market will be multiple DSOs, multiple

4.2 Background knowledge of the market based control solution 43

easier for the DSO to use the price to regulate the FO’s energy consump-tions.

• Considering the fact that DSOs can be seen as a monopoly inside its net-work, it is reasonable for the distribution system operator being a centralprice coordinator whose role is to coordinate the load behaviors of the com-mercial actors. The central price coordinator is supported by the generalequilibrium market mechanisms.

• Comparing the general equilibrium market mechanisms and the doubleauction approach, one simple explanation is that the former uses priceto penalize the power consumptions during the peak moments, while thelatter regards price as an incentive to shift the power consumptions duringpeak time. If the size of the market is not flourishing, it is assumedthat the former one will solve the grid congestion more efficiently. Morediscussion regarding the comparisons between the ‘penalty scheme’ andthe ‘compensation scheme’ is presented in the discussion section of thischapter, i.e., section 4.4.

• The general equilibrium can be easily adopted for large scale of distributedenergy resources, individual households supported by the recently devel-oped powerful computational devices [89] and the idea is recently studiedin [90] where home area network management system and distributionutility company exchange information on energy consumption and price.

4.2.2 General equilibrium market mechanisms

General equilibrium market mechanisms is a microeconomic market frameworkthat has been recently been successfully adapted for and used in computationalmulti-agent systems in many application domains [86, 87, 91]. Some propertiesof general equilibrium such as Pareto efficiency, coalitional stability, existence,uniqueness under gross substitutes, and convergence have been discussed in[91]. It is pointed out that there are many algorithms which can be used tosearch for a general equilibrium, some centralized, and some decentralized. Themost common decentralized algorithm for this purpose is the price tatonnementprocess which is a steepest descent search method. The pseudo-code of thealgorithm is presented in [91]. The distributed price tatonnement algorithm ismodified in the present thesis to facilitate the allocation of the grid capacity tothe FOs. The algorithm can be mathematically supported by the combinationof dual decomposition method and subgradient method which are described in

FOs or aggregators which means the number of the actors will be countable. Besides, notethat the purpose of this market is to solve the grid congestion issue, which means for eachDSO, they only interact with the FOs who use its network.


section B.5. In the process of find the equilibrium, the market participant, i.e.,the fleet operator can choose different approaches to participate into the market,such as strident antagonist or cooperative antagonist, described in section B.6.It is assumed that the market participant will honestly behave in the market inthe present study.

4.3 Main results

4.3.1 Map of the market based control operation

As we have explained in chapter 1, four functional levels are applied to addressthe congestion problems and have been adopted into four fundamental stages:1, Offline planning. 2, Online scheduling. 3, Real time operation. 4, Offlinesettlement.

EV Owner

DSO

FO

II. Scheduling III. Operation IV. Settlement

Driving pattern

prediction

Online grid

measurements

SLA: EV Owner – FO terms

of charging management

targeted

Real-time

interventions

EV utilization

(driving and charging)

Energy market

settlement

Stage

Sta

keh

old

er

I. Planning

Notification of non-typical

driving needs.

Selection of market

products and services Grid overload

forecasting

Market/service

forecasting

Dimensioning

of Grid Fuses and relay operation Basic

Advanced

automationGrid state

estimation

Smart Meter

data collection

and aggregation

Grid Load

forecasting

(annual)

Meter data

collection

Load Curve

EstimationFault Analysis

Repairs

Charging Control

Demand

forecasting

EV owner settlement

Ancillary Service

control signals

Optimal

charging plan

calculation

Volt/Var

optimization

Reconfiguration /

Adjustment of transformer taps

Market contracting

Iterative calculation of

market prices

Figure 4.1: Map of distribution grid capacity market operation

In the first stage, planning has been distinguished from scheduling in the samefashion as unit commitment is distinguished from dispatch: depending on thespecific coordination strategy, we distinguish operations that can be coordinatedin a ad-hoc fashion and those that provide the basis for such ad-hoc decisions.The online scheduling stage can in time be closely coupled with operation (e.g.

4.3 Main results 45

reactive scheduling with a 5min resolution) or extend hours or days ahead ofit. Scheduling is the stage in which available resources are best known andthe platform for execution is to be prepared. The operation stage is aboutpure execution in real-time. Plans are only executed, and unplanned eventsoccur and physical as well as automatic controls respond without deliberation.Settlement is about the aftermath: recordings (measurements, sent commands,etc.) of executed operations are consolidated and (financial) responsibility isallocated. Due to these clear distinctions, the four stages supports the discussionof interactions between key operation tasks for cross-stakeholder coordinationfor the complete process.

In the study, the operations DSO, FO and EV owner would be required toexecute in the market based control scheme are mapped out which is shown infig. 4.1. More explanations are found in paper A.5.

4.3.2 Market based approach for grid congestion preven-tion

We focus on stage 2, i.e., scheduling and the main results of developing themarket based control for grid congestion prevention are presented in paper A.6.A cost and schedule adjustment algorithms modified from previous introduceddistributed price tatonnement algorithm is presented in the following. Thepurpose of the algorithm is to minimize the charging cost of EV fleet as wellas preventing the grid congestions. The above algorithms describe the detailsof the market based control. For case studies and simulations, please refer topaper A.6.

4.3.3 Price coordinated hierarchical scheduling system

Based on the understanding obtained from the study in paper A.5 and A.6,the proposed scheme is further extended into a price coordinated hierarchicalscheduling system in paper A.7. The extensions lie in the discussion of applyingseveral approaches to find the price inside the market, such as uniform price auc-tion mechanism and the general equilibrium market mechanisms. In addition,we also discuss the control relationship between the FOs and the EVs, such asinstead of central scheduling and control of the charging profile of the EV fleet,the EV users can individually make the charging schedule given the price. Thecombination of applying market based control for the interaction between theDSO and the FOs and using the decentralized coordination between the FOsand the EVs is simulated and presented in paper A.9.


Algorithm 1 Cost and schedule adjustment algorithm

function 1. Schedule generation and congestion verificationFOs calculate the optimal charging schedule and send the schedule to the

DSO/market operator.The DSO/market operator conducts congestion verfication, if happens, go

to function 2, otherwise, the energy schedule is approved.end functionfunction 2. Operation of Market based control

Initialize the shadow price which reflects the grid capacity margin.loop

DSO/Market operator sends the price to all the FOs.FOs updates the charging schedule and bid the charging schedule to the

market operator.The market operator updates the shadow price and broadcasts it to all

the FOs, until the price converges.end loop

end functionfunction 3. Adjustment

loopThe new shadow price is sent to FOs.Go through function 1 and 2, until Λ(i) ≤ 0, i.e., the shadow price is

less than or equal to zero, then terminate the iteration.end loopBid final energy/power schedule to the electricity spot market.

end function

In paper A.9, we compare the results of two cases where the DSO both usethe market based control to interact with the FOs, however, the coordinationmethods between the FOs and the EVs are different. In the first case, three fleetoperators are assumed to centrally schedule and control the electric vehicle’charging which is also the scenario in paper A.6, while in the second case,it is assumed that three fleet operators only aggregate the charging scheduleswhich are made by the electric vehicle controllers. The results show that thecongestion problems are solved after 5 steps in the first case while only 2 stepsin the second case. The difference is because that the electric vehicles in the firstcase are always responding the shadow price and trying to avoid the chargingon the higher price period, as a result, the electric vehicles will be scheduledto charge at other lower price period where congestion might happens as well.While in the second case, only some electric vehicles are assumed to responds tothe shadow price which means that only part of the charging plan is rescheduledto other lower price period and thereby reduce the possibility of causing a newcongestion period.

4.3 Main results 47

4.3.4 A flexibility clearinghouse concept (FLECH)

As aforementioned that this project is associated with the Danish iPower project,one of the highlights of the project is the concept of FLECH (a flexibility clear-inghouse). In paper A.8, we presented the concept for facilitating ancillaryservices at the distribution system level. With the emergence of new play-ers in distribution system ancillary service markets, it is foreseen that such amechanism will be needed to minimize transaction costs. In contrast to othercontributions on distribution congestion mitigation, the FLECH adapts to theactual DSO needs and is not tied to a specific aggregator architecture. The roleof FLECH and its interactions with stakeholders of the distribution flexibilityservice market are illustrated in Fig. 4.2. Note that the single side auctionmechanism is used by the FLECH to find the price, i.e., the equilibrium in themarket.

Flexible Demand (DER)

Electricity

Spot Market

Grid

Measurements

Fleet Operator

Distribution

Operation

co

ntro

l

Information

flow

Control

relation

Conventional

Demand

Ancillary

Services

Physical

connection

Aggregation

Smart

Meter

Data

DER

Owner

Distribution

system

operator

Transmission

System

Operator

Other market

participants

Fleet

operator

FLECH

FLECH

operator

Figure 4.2: Schematic overview of the considered actors and their roles in rela-tion to the FLECH

A case study has been presented showing how FLECH is envisioned to facilitatethe service required by the DSO. In the scenario, distribution system operatorforesees that a low voltage transformer will be overload by the power consump-tion of electric vehicles.


At the planning stage, the DSO submits the following service tender to FLECH:PowerMax:Ensure that the capacity limits specified here are not violated :

CAPACITY REDUCTION [AREA T1]: 37.8kW

TIME 4:30pm TO 8:00pm ON weekdays

PERIOD 01 Dec 2014 TO 28 Feb 2015

RECOMMENDED RATE 10 EUR/kW

This tender is then announced by FLECH to all fleet operators registered forarea T1. The fleet operators bid into the FLECH:AggID [BidID]: reduction FROM capacity AT rate flex?

FO1[FO1B1]: 12.3kW FROM 14.8kW AT 10 EUR/kW FULL


FO2[FO2B2]: 15.4kW FROM 18.5kW 20 EUR/kW FLEX

Note that FO1 did not bid with all of its resources, effectively only using 4 outof 5 cars, and that the second bid by FO2 is FLEX bid, i.e., it does not needto be accepted entirely. After gate closure their bids are forwarded to the DSOwhich evaluates the offers and decides to accept the following bids:BidID: FO1B1, FO2B1, FO2B2*90% AT 20 EUR/kW

This leads to an effective capacity reduction of 38.5kW which fulfills the required37.8kW . The prices of this case study are completely fictitious and not anchoredin real costs. We further use a multiagent system to demonstrate the case study,the developed multiagent system will be described in section 5.2.2 of Chapter 5.


4.4.1 Comparisons among the three control strategies

Although the centralized control and price control for distribution grid conges-tion management are not investigated directly, some comparisons can still beperformed thanks to relevant topics studied in [9] and [8]. In both studies, gridcongestion management and EV integration are investigated. In [9], day-aheaddynamic tariff control is used. In [8], centralized approach is utilized. In order tointuitively illustrate the interactions/control relationships between the involvedactors, Fig. 4.3 shows information flows of the market based control, pricecontrol and centralized control for distribution grid congestion prevention. Weadapt and draw the information flows presented in [9] and [8] into the currentone.


1. EV driver selects charging requirements, EV status

FOs

2.EV information to FO

Public network

Distribution system

operator

Distribution grid Market

operator

3. FO (Re)Generates the charging schedule and

submits the aggregated charge schedule to DSO

4. DSO verifies charging schedule of Fos.

5. DO submits the charge schedule to

market operator

6. Market operator sends the shadow price to Fo

EV owner


FOs

Public network

Distribution system

operator

Spot Market

operator

3. FOs Sends a price reflecting the forecasted state of distribution network

4. Fos generate the charging schedue considering the

price sent by DSO and submits the charging

schedule (bids) to spot market operator.

EV owner

Price control (Day-ahead

dynamic tariff based control)

FOs

Public network

Distribution system

operator

Spot Market

operator

3. FOs implement charging schedule optimization and

send the preferred load curves to DSO

4. DSO verifies charging schedule And set grid

constraints

5. DO submits the charge schedule to market operator

EV owner

Centralized approach





Spot Market

operator

7. Fos bid into the market.

(a)

(b)

(c)

Figure 4.3: Information flow of the three control strategies for distribution gridcongestion prevention

From figure 4.3, it is shown that EV owners firstly communicate with the FOto define their personal requirements in all three studies [8, 9] and paper A.6.Then, the FOs make the charging schedules for the electric vehicles. In thecase of price control, the FOs may be firstly notified by the DSO with a pricesignal reflecting the forecasted congestions, then the FOs generate the chargingschedule considering the price sent by the DSO and submit bids to the spot


market. This is different with the market based control and centralized controlmethod where the FOs need to coordinate with the distribution grid marketoperator and DSO to prevent the congestions. The interactions between the FOand the DSO/market operator may require multiple iterations. Note that inthe centralized control, the FOs will get an constraint (power limits) from theDSO every iteration, it is not addressed in [8] that how the constraint is set foreach FO if multiple FOs share one power distribution network. In the marketbased control, this is handled by the economic principles, and the negotiation isfacilitated by a market operator. It should be pointed out the market operatorcould be an independent entity such as the case in A.8 or be the same entitywith the DSO such as the case in paper A.6.

Some comparisons are listed in Table 4.1, which is shown in the following.

Table 4.1: Comparison of control strategies for distribution grid congestionmanagementActor Complexity Value Risk Other issues

Market based controlFO High High Low Privacy improved, flexible, price

and technical constraint are re-solved via the general equilib-rium market mechanism.

DSO Relativelyhigh

High Low Enable a comparatively high uti-lization factor of the grid.

Price controlFO Medium Low Low Easy to interact, price received.DSO Medium Medium High N.A.

Centralized controlFO Medium Low Low Easy to interact, technical con-

straint received.DSO Medium Medium Low N.A.

The overall rationale is to provide cost-efficient solutions, no firm quantificationof benefits or costs have been performed in the thesis. In addition to this, thisthesis has not addressed one aspect of distribution grid congestion management,i.e., the long-term problem of distribution grid reinforcement-deciding when andwhere to build new transmission facilities. Such discussions can also be foundin [43,52] where the importance is emphasized and some illustrative calculationsare presented. However, it is clearly stated in [43,52] that smart control strate-gies incurred demand response is most likely the attractive way for the DSO tosolve the congestion if it only occurs a few times per year. Besides, regardingthe discussion on pricing through compensation mechanism or through penalty


mechanism, if we use a dialectic perspective to view the two mechanisms, thereis no big differences. For example, the cost saved by the DSO on new investmentcould be used for compensation in the congestion market or in another way, thesaving could be directly share with all the customers firstly, then, the DSO canuse the price to penalize the one who use the electricity in the peak time.

4.4.2 Market based control related issue

The overall assumption for market based control is that there exists an equi-librium price that will clear the negotiation in each round, e.g., the negotiationbetween the market operator and the fleet operators. This works fine whenthe cost functions of the fleet operators are convex. However, the presence ofunits cost functions that are not strictly convex which brings some difficultiesin finding the clearing price. In a traditional auction-based market, like Elspotor the regulating power market when the price is determined by submitted bids,this problem can be overcome by only accepting the bids needed. However, inthe present case, the market operation method is different. In the present case,one situation might happen, i.e., that the fleet operators respond the same wayif the given price range is not obvious which means that the iteration numbersmay increase a lot to find the clearing price. Some studies [90, 92, 93] havebeen given on solving this problems. In [90, 93] the suggestion is using primalaveraging method if the objective function of the market participant is neitherstrictly convex nor finite and the authors in [93] proved that the algorithm (dualdecomposition and sub-gradient method which supports the price finding in thegeneral equilibrium market) finds near-optimal schedules even when advancedmetering infrastructures (AMI) messages (updated prices and residential load)are lost, which can happen in the present communications network. In [92], amethod using randomized price offsets was presented to deal with non-convex,non-continuous supply and demand functions. The method relied solely onsmall but fair alternations of the price signal for each prosumer leaving theutility functions untouched and the results showed that both the convergencebehavior and solution quality improves with the market size

Besides, although using market based approach has been demonstrated to en-able an optimal resource allocation, the uncertain in terms clear business modelsfor fleet operators and distribution system operators, and the necessity for theregulatory support makes using price as a coordination tool for serving grid ser-vices is still a challenging task. But, some lessons can be drawn from the presentstudy, if the size of the market is great, an independent capacity market couldbe created to facilitate the transactions between distribution system operatorsand fleet operators, maybe in a national level. However, if the size of the marketis small, the market based control could be facilitated by a functions insides the

52

distribution system operation. Nevertheless, to ensure enough competition andfairness of the capacity market, one prerequisite is the number of market par-ticipants, i.e., fleet operators. If there are few fleet operators in the distributionarea, issue e.g., market power will become a major challenge from the marketperspective. A mechanism needed to be developed to hedge the risks.

Chapter 5

Multi-agent system fordistribution grid congestion

management

In this chapter the main results regarding the third research objective are sum-marized. The main results have been mainly presented in the manuscript A.9and A.10, included in appendices of this report. The chapter starts with ageneral discussion on using multi-agent system (MAS) for a hierarchical organi-zation based distribution systems. Then, the main finding of the third researchtopic are summarized.

5.1 Characteristics of multi-agents system

Jennings and Bussmann [94] pointed out that modern control systems must meetincreasingly demanding requirements stemming from the need to cope with sig-nificant degrees of uncertainty, as well as with more dynamic environments, andto provide greater flexibility. This, in turn, means that control systems soft-ware is highly complex in that it invariably has a large number of interactingparts. It is argued in the article that analyzing, designing, and implementingsuch complex software systems could be implemented as a collection of inter-acting, autonomous, flexible components (i.e., agents) which affords softwareengineers several significant advantages over contemporary methods. The studyfirstly discussed that a canonical view of a complex system can be defined asthe following in Fig. 5.1. The system’s hierarchical nature is expressed through

54 Multi-agent system for distribution grid congestion management

the “related to” links; components within a subsystem are connected through“frequent interaction links”, and interactions between components are expressedthrough “infrequent interaction” links. The complex system specifically in thearticle is complex software systems (of which control systems is an instance).

Subsystem

Subsystem component

Frequent interaction

Infrequent interaction

Related to

Figure 5.1: View of a complex system

It is further discussed that software engineers have devised several fundamentaltools to help manage this complexity including decomposition, abstraction andorganization approaches. Then the authors [94] argued that the case for agentoriented software engineering. The arguments are mainly performed from threeaspects: 1) The merits of agent-oriented decompositions. 2) The suitability ofagent-oriented abstractions. 3) The need for flexible management of changingorganizational structures. In the article, a canonical view of an agent-basedsystem (Fig. 5.2) is presented. Fig. 5.2 shows that: 1) adopting an agent-oriented approach to software engineering means decomposing the problem intomultiple autonomous components that can act and interact in flexible ways toachieve their set objectives; 2) the key abstraction models that define the agent-oriented concepts are agents, interactions, and organizations; and 3) explicitstructures and mechanisms are often used to describe and manage the complexand changing web of organizational relationships that exist between the agents.Regarding organizational relationships, the author in [95] show that four orga-nizations are widely used for computer-based agents. Organization means a setof policies (rules and criteria) by which agents are aggregated to form greateror super-agents, just as birds, fish and people aggregate to form flocks, schoolsand corporations. In the study, it is discussed that a data flow and a controlflow describe the structure of a super agent. Furthermore, the data flow is par-titioned into two fuzzy subsets: the subset of strongly cyclic data flows, and its

5.1 Characteristics of multi-agents system 55

complement; the control flow is partitioned into two fuzzy subsets: the subsetof hierarchical control flows and its complement. Then, the super-agent can bepartitioned into four regions, i.e., cyclic data, hierarchical control; cyclic data,null control; acyclic data, hierarchical control; acyclic data, null control.

Physical Environment

Agents

Interaction

Organizational

relationship

Physical environment

Physical device

Related to

Figure 5.2: Canonical view of an agent-based system interacting with a physicalenvironment

When adopting an agent-oriented view of the system, it is apparent that theagents will need to interact with one another, either to achieve their individualobjectives or to manage the dependencies in the common environment. Theseinteractions can vary from simple semantic interoperation (information pass-ing), through traditional client-server-type interactions, to rich social interac-tions (the ability to cooperate, coordinate, and negotiate about a course of ac-tion). FIPA (The Foundation for Intelligent Physical agents) standards 1 definea framework for inter-agent interaction in multi-agents system, it also speci-fies abstract architectures for multi-agents system implementation, as shown inFigure. 5.3.

Besides the application of agents, the FIPA architecture includes several ser-vices. A Directory Facilitator provides yellow page services i.e., different agentsinteract with this service to register and discover available agent services. AgentManagement Services provide white page services. This agent is responsible forcreating, destroying and managing agents and containers in a multi-agent plat-

1FIPA, The Foundation for Intelligent Physical Agents:http://www.fipa.org/


Figure 5.3: FIPA abstract architecture for MAS implementation

form. The Message Transport Service is responsible for message transportationbetween agents. This service also enables synchronization of messages whenseveral messages are sent and received from different agents in parallel. FIPAalso provides a language specification for communication between agents. Thislanguage specification is called the FIPA agent communication language (FIPAACL).

5.2 Main results

5.2.1 Modeling the price coordinated hierarchical schedul-ing system by multi-agent concepts

In previous chapter, a price coordinated hierarchical scheduling system is pro-posed to optimally integrate electric vehicles into the distribution network wherethe DSO uses market based control to allocate the grid capacity among the FOs,and the control policies between the FO and the EVs are implemented bothcentrally and distributed. To simulate and evaluate the proposed scheme, amulti-agent system based technology is very suitable for modeling the proposedcontrol systems. This can be justified by the following reasons:

• The increase in complexity and size of the whole electric vehicle chargingnetwork bring up the need for distributed intelligence and local solution,which fall into the scope of MAS based technology.

• The information flow, optimizations and the negotiations in the smartcharging network of electric vehicles can be well demonstrated and inte-

5.2 Main results 57

grated into a MAS.

• The system can be pre-tested and pre-analyzed by using a MAS beforegoing to real implementation.

• The MAS can be used to test different control strategies, various businessscenarios, and different behaviors of the market participants.

An agent view of the price coordinated hierarchical scheduling system is pre-sented in Fig. 5.4. In the figure, operations of the agents are not shown,nevertheless, it is shown that multi-agent based technology can be used to char-acterize a complex hierarchical system for scheduling.

Fleet operator agent layer with

agent directly Schedule/indirectly

coordinate the charging behavior

of Evs.

Physical Power Systems layer with devices such as

conventional loads, Electric vehicles, and heat pumps etc.

Central control agents Layer with

implementation of higher level of

control and communication with

medium level agents

EV agent level with agents

represent the EV operation

and interact with the up-level

FO agent.

Figure 5.4: Agents view of the price coordinated hierarchical scheduling system

5.2.2 Platforms Development

To implement the multiagent systems described above, Fig. 5.5 illustrates theMAS system architecture, in which, all the agents are built in JACK which is


FOi agent

DSO agentEV agent-n

Market operator

agentMulti Agents

coordination layerJACK

MULTIAGENTS SYSTEM

FOi

optimization slover for market

participationMatlab based

EViSchedule

generationMatlab based

Conventional load

EVn

Market operation

Matlab based

Grid SimulationSmulink

Optimization tools Matlab

Figure 5.5: A MultiAgent system architecture.

an agent-oriented development environment built on top of and fully integratedwith the Java programming language [96]. JACK offers the environment andfacilities for message sending/receiving. MATLAB based functions enables aprocedural implementation of decision modules. Simulink is used to model thedistribution grid and for power flow calculation. The java application program-ming interface matlabcontrol 2 is used for JACK to interact with MATLAB.DSO agent interacts with the Simulink through a matlab file. A case studyis performed to examine the developed platform. In the case study, one DSOagent, one market operator agent, three FO agents, and fourteen EVs agents areincluded. The simulations show that it well emulates the negotiation behaviorinside a capacity market by the interaction diagrams. For more details such asthe features of JACK, how the market actors in the system are mapped intothe agents in the multiagent systems, and the simulation results, the readers arereferred to manuscript A.9.

The platform has also been used to simulate a negotiation bidding frameworkin [97]. The negotiation sequence in [97] following the discussions in paperA.8 where three different agents, the DSO, the FLECH and the Fleet operatoragent aim to eliminate the grid congestions by market based approach. Thenegotiation sequence shall find a price, i.e., the equilibrium for the market ne-

2https://code.google.com/p/matlabcontrol/


gotiation when services are initiated by the DSO. The framework is focusing onthe day-ahead market and is using an auctioning approach to clear the market.It is assumed that this clearing does not necessarily happens in first biddinground, but could extend to several bidding rounds before the market clearanceis found, although a max of 7 rounds is implemented. Several case studies havebeen tested to show the functionality of the negotiation platform including asimple case clearing in first round, two cases need multiple rounds where priceand power are regulated to find the equilibrium, and a extreme case where theprice is too high to find the equilibrium. The case studies illustrate that theplatform is able to conduct negotiations which lowers the fleet operators’ con-sumption in a certain period in exchange of an economic compensation to thefleet operators.

In addition, it is planned in paper A.10 that the platform will be used to explorea case study where the fleet operators have slightly more complex operations.In the study, the multiple fleet operators will do the similar negotiations withthe distribution system operator as the one shown in paper A.9, but the fleetoperators in paper A.10 will try to explore more economical benefits whichcan be offered by the electric vehicles. These extra economical benefits willbe achieved by using the EV’s discharging capability. The question is formu-lated as a mixed-integer non-linear programming based problem, implementedin General Algebraic Modeling System (GAMS) 3 using the DICOP Solver 4.


In this study, we developed and used an integrated environment consisting ofJACK agent software and MATLAB/Simulink to analyze the interactions be-tween the agents considering their own decision making, information flow ex-changes and the influences on the physical grids. JACK is good for demon-strating the coordination schemes among the actors, MATLAB is competent tocarry out the technical computation of optimization problem, Simulink is knownfor its interaction with MATLAB and it is adequate to perform the technicalvalidation for the present study. In a general case, various simulation platformscan be utilized in a distribution grid congestion demonstration. For example,besides JACK, JADE 5 is also widely used for multiagent simulation. JACKis chosen because of its capability and support for the explicit modelling of thetypical MAS entities such as agent, plan, event and capabilities. Moreover, inJACK it is easier to design and analyze interactions and dependencies among

3http://www.gams.com/dd/docs/bigdocs/GAMSUsersGuide.pdf4http://www.gams.com/dd/docs/solvers/dicopt.pdf5http://jade.tilab.com/

60

such entities. In term of solving an optimization problem, GAMS (the GeneralAlgebraic Modeling System) 6 also has good performance, however, MATLABis easier to be used and seems more popular in the academically field. Last butnot least, grid modeling tool is also an important part. The currently existinggrid modeling tools include Simulink, MatPower 7, PowerFactory 8, ARISTO 9,NEPLAN 10 etc. However, here it is not the aim to give a comparison of thevarious platforms, instead, we want to emphasize that the relevant tools can beintegrated with the MAS settings.

Currently, the agents in this project can be categorized as being reactive be-cause the agents chooses their actions based solely on the immediate input andsome logical control defined for the study. In order to fully use the advantagesof different simulation tools, especially if the multi-agent system is used forreal time control simulation, a simulation methodology such as using high-levelarchitecture that allows individual subsystems/components to be simulated bydifferent simulation tools running simultaneously and exchanging informationin a collaborative manner need to be implemented. This could be a topic forfuture work.

6http://www.gams.com/7http://www.pserc.cornell.edu//matpower/8http://www.digsilent.de/index.php/products-powerfactory.html9Advanced Real-Time Interactive Simulator for Training and Operation, Sweden

10http://www.neplan.ch/

Chapter 6

Conclusions and Future Work

This thesis firstly examined three control strategies for the exploitation of servicebased electric vehicle aggregation, mainly from a commercial actor’s perspective.Then, the thesis investigated the control strategies for the distribution systemoperator to coordinate with the commercial actors to eliminate grid congestion.Finally, to evaluate the proposed control strategies, a multi-agent system is usedto model and simulate the system. The main outcome of this project is a pricecoordinated hierarchical scheduling system for large scale integration of electricvehicles into the power distribution network. In this system, the distributionsystem operator uses market based control to interact with the commercial actorto prevent the grid congestions. The commercial actor uses either direct controlor price control to manage the distributed charging behavior of electric vehiclesdepending on the actual situations. This chapter presents a number of keyresults from the present study and recommends several topics for future work.

6.1 Conclusions

In section 1.2.2, three research problems are outlined for the project. In thefollowing, the results are concluded:

1. For a commercial actor (fleet operator in the present thesis), how to choosea control strategy to manage a large scale of electric vehicles?

Based on the review and methods development presented in chapter 3, cen-tralized direct control and price control are concluded to be more suitable

62 Conclusions and Future Work

comparing to market based control for management of a EV fleet. Cen-tralized direct control offers the best performance in controlling the EVfleet and therefore making better optimized charging profiles, especially ifthe participation into the regulating power markets was considered. Forthe centralized direct control, it is recommended that linear programmingbased techniques is considered to characterize the optimal charging prob-lem and generate the optimal charging schedule. In order to generate anoptimal charging schedule, a forecasted electricity price and predicted EVsdriving pattern are essential. Fortunately, they can be estimated by thecommercial actors. Nevertheless, EV owners are encouraged to submit aprovisional EV utilization plan for next day to the commercial actor forgenerating an optimal charging schedule. For the centralized control tobecome a success, more research is needed in setting up a collaborativebusiness model which ensures the proper engagement of commercial actorsand EV owners.

Price control is probably the most attractive way for the commercial ac-tor to regulate the charging behavior of the electric vehicles considering itseasier implementation. It is especially effective in the case of decreasingthe charging in the peak time for DSO or the case of increasing the load fortransmission system operator 1. For the price control to become a success,more research is needed in price responsive models or price elasticity mod-els to obtain satisfactory performance and grid reliability. Market basedcontrol is not recommended here considering some barriers, such as anautomated negotiation device need to be mounted in the electric vehiclewhich performs the enabling function if using the market based control.Such automated control devices are not currently available.

2. To manage the congestion of the distribution network, which control strat-egy should be considered taking into account the distribution system op-erator, the commercial actor’s self-interests and operational constraints?

The thesis focused on the development of control strategies for the DSOto interact with the commercial actors instead of distributed demand sideunits such as individual EVs, heat pumps etc. Three control strategieshave been examined and compared in section 4.4.1. The chosen marketbased control is being developed and tested. There are several advantagesfor the utilization of market based control. First, market based control isproven to be an effective way of allocating conflicting resources. Second,the negotiations required in the market can be enabled and operated bythe DSO and commercial actors without much burden. Third, market

1Using price control for providing services to transmission system operator is not the fo-cus of this thesis, for such discussion, please refer to Ecogrid EU project http://www.eu-ecogrid.net/. In addition to this, in [98], the authors prioritizes the needs to handle theconflicting resources, e.g., the services required by the transmission system operator mightcause overloading issue for distribution network.

6.2 Future Work 63

based control can ensure a low risk for DSO when solving the grid con-gestion, since price and power schedule are resolved via the price clearingmechanisms. In section 4.2.2, general equilibrium market mechanisms areused to find the clearing price instead of auction approach. In order tobenefit from the market based control strategy, more work is needed forstudying the non-convex cost problem which has not been addressed inthis thesis.

3. Development of tools to simulate the control strategies developed for dis-tribution grid congestion management.

It is demonstrated that multi-agent system based technology is very suit-able to evaluate the development of market based control strategies con-sidering the DSO and FO’s operations. Motivations have been given insection 5.2.1, in addition, for a practical case using the market based con-trol scheme, the market participant may behave differently, which can alsobe well simulated in a multi-agent system.

6.2 Future Work

The present research addressed a comparatively open and new field of study andthere is a large scope for continuing the current work. Some topics are listedbelow to direct future research as a follow-up to this project.

• Commercial actors, having integrated control strategies such as direct con-trol and price control may be a feasible solution to mitigate the downsideof the price control. In addition, considering the fact that maybe somecustomers prefer their EVs directly controlled while some EV users wouldlike to receive the price signals and respond the price individually, it isinteresting to make a study to analyze the related issues. A preliminarystudy [99] has been performed where the modern portfolio theory wasadopted.

• Similarly, the distribution system operator, having an integrated controlscheme such as direct control, market based control and price control maybe a feasible solution to mitigate the uncertainty of the price control, thehigh communication cost of market based control and the high computa-tional cost of direct control. It is interesting to e.g., design a frameworkfor choosing the suitable control strategies such as one that is based on theavailable controllable resources, the control requirements, and the relevantinformation and communication facilities.

64 Bibliography

• The overall rationale of the present work is to provide cost-efficient solu-tions, but no quantification of benefits or costs has been performed. Futureanalyses could be performed on the cost of the different control concepts.For example, for an advanced control concept, the immediate benefits mayappear less obvious, compared with the complexity that is introduced torealize it. In addition, the long-term marginal costs of reinforcement ofthe grid could also be performed in order to be able to compare these costswith the cost involved to make the end-users reducing demand in a givenperiod to avoid congestions in the distribution grid.

• In order to fully understand the grid congestion problem, a distributiongrid power flow calculation method should be chosen and used to supportthe control strategy. For example, the network flow model and Newton-Raphson method have been used in [8] where centralized control is usedto prevent the grid congestion; the optimal power flow based calculationis used in [9] where day-ahead dynamic price control is used to managethe grid congestion. To fully integrate the market based control withthe power flow calculations, the recently proposed alternating directionmethod of multipliers (ADMM) could be used to facilitate the integration.This idea is currently under investigation, please see the further discussionin section B.7.

• Mainly, thermal limits of the distribution grid has been investigated in thepresent work, but it is also known that other constraints such as voltagelimits are important and need to be studied. In a practical way, in theplanning period, the DSO considers and pre-handles voltage problem byreinforcing the grid infrastructure based on the regulations. These regu-lations describe the allowed voltage safety limits in the distribution grid.In the normal operation period, on the substation level, transformers hastap changers which can be used to regulate the voltage. In Denmark, ingeneral, 60kV/10kV transformer has on-load tap-changers. In the future,the system operator could set up grid codes for DERs, requiring the DERsto have their own embedded voltage control, which could solve the prob-lem preventively. In the context of this study, voltage control can alsobe implemented by e.g., market based control scheme which could be aninteresting topic for future work.

Bibliography

[1] W. Justin and M. Jacopo, “Wind in power: 2012 european statistics,”The European Wind energy association, Tech. Rep., 2013. 1.1

[2] E. DK and D. E. Association, “Smart grid in denmark 2.0,” Energinet.DK and Danish Energy Association, Tech. Rep., 2013. 1.1, 4.1.1

[3] L. E. Jones, “Strategies and decision support systems for integrating vari-able energy resources in control centers for reliable grid operations,” Al-stom Grid Inc., Tech. Rep., 2012. 1.1

[4] C. Binding, D. Gantenbein, B. Jansen, O. Sundstrom, P. Andersen,F. Marra, B. Poulsen, and C. Traeholt, “Electric vehicle fleet integra-tion in the danish edison project - a virtual power plant on the island ofbornholm,” in Power and Energy Society General Meeting, 2010 IEEE,2010, pp. 1–8. 1.1

[5] B. Jansen, C. Binding, O. Sundstrom, and D. Gantenbein, “Architectureand communication of an electric vehicle virtual power plant,” in SmartGrid Communications (SmartGridComm), 2010 First IEEE InternationalConference on, 2010, pp. 149–154. 1.1

[6] P. B. Andersen, “Intelligent electric vehicle integration-domain interfacesand supporting informatics,” Ph.D. dissertation, Technical University ofDenmark, 2012. 1.1

[7] O. Sundstrom and C. Binding, “Planning electric-drive vehicle chargingunder constrained grid conditions,” in 2010 International Conference onPower System Technology (POWERCON),. IEEE, pp. 1–6. 1.1, 2.2

66 BIBLIOGRAPHY

[8] ——, “Flexible Charging Optimization for Electric Vehicles ConsideringDistribution Grid Constraints,” IEEE Transactions on Smart Grid, vol. 3,no. 1, pp. 26–37, 2012. 1.1, 2.1, 2.2, 2.4, 2.2, 3.1.2, 3.1.3, 4.4.1, 4.4.1, 6.2,B.1, B.1

[9] N. O’Connell, Q. Wu, J. Østergaard, A. Nielsen, S.-T. Cha, and Y. Ding,“Electric vehicle (ev) charging management with dynamic distributionsystem tariff,” in 2011 2nd IEEE PES on Innovative Smart Grid Tech-nologies, 2011, pp. 1–7. 1.1, 2.2, 4.4.1, 4.4.1, 6.2

[10] D. Popovic and V. Bhatkar, Distributed computer control for industrialautomation. CRC, 1990. 1.2.2

[11] W. Kempton and S. E. Letendre, “Electric vehicles as a new power sourcefor electric utilities,” Transportation Research Part D: Transport and En-vironment, vol. 2, no. 3, pp. 157–175, 1997. 2.1, 2.2, 2.1

[12] W. Kempton and J. Tomic, “Vehicle-to-grid power fundamentals: Cal-culating capacity and net revenue,” Journal of Power Sources, vol. 144,no. 1, pp. 268–279, 2005. 2.1, 2.1, 2.1

[13] ——, “Vehicle-to-grid power implementation: From stabilizing the grid tosupporting large-scale renewable energy,” Journal of Power Sources, vol.144, no. 1, pp. 280–294, 2005. 2.1, 2.1

[14] J. Tomic and W. Kempton, “Using fleets of electric-drive vehicles for gridsupport,” Journal of Power Sources, vol. 168, no. 2, pp. 459 – 468, 2007.2.1, 2.1

[15] C. K. Divya, J. Østergaard, E. Larsen, C. Kern, T. Wittmann, andM. Weinhold, “Integration of electric drive vehicles in the Danish electric-ity network with high wind power penetration,” European Transactionson Electrical Power, vol. 20, no. 7, pp. 872–883, 2010. 2.1

[16] J. Østergaard, A. Foosnæ s, Z. Xu, T. A. Mondorf, C. A. Andersen,S. Holthusen, T. Holm, M. F. Bendtsen, and K. Behnke, “Electric ve-hicles in power systems with 50% wind power penetration: the Danishcase and the EDISON programme,” in European Conference Electricityand Mobility. 2.1

[17] J. Østergaard and J. E. Nielsen, “The Bornholm Power System-Anoverview,” 2008. 2.1

[18] E. Larsen, D. Chandrashekhara, and J. Ostergard, “Electric vehicles forimproved operation of power systems with high wind power penetration,”in Energy 2030 Conference, 2008. ENERGY 2008. IEEE, 2008, pp. 1–6.2.1

BIBLIOGRAPHY 67

[19] C. Guille and G. Gross, “A conceptual framework for the vehicle-to-grid(v2g) implementation,” Energy Policy, vol. 37, no. 11, pp. 4379 – 4390,2009. 2.1

[20] N. Rotering and M. Ilic, “Optimal Charge Control of Plug-In HybridElectric Vehicles In Deregulated Electricity Markets,” IEEE Transactionson Power Sy, 2010. 2.1, 3.1.2, 3.1.3, B.1, B.1, B.1

[21] S. Han and K. Sezaki, “Development of an optimal vehicle-to-grid aggre-gator for frequency regulation,” IEEE Transactions on Smart Grid, vol. 1,no. 1, pp. 65–72, 2010. 2.1

[22] J. Pillai and B. Bak-Jensen, “Integration of vehicle-to-grid in the westerndanish power system,” Sustainable Energy, IEEE Transactions on, vol. 2,no. 1, pp. 12–19, 2011. 2.1

[23] J. A. Pecas Lopes, P. M. Rocha Almeida, and F. J. Soares, “Using vehicle-to-grid to maximize the integration of intermittent renewable energy re-sources in islanded electric grids,” in Clean Electrical Power, 2009 Inter-national Conference on. IEEE, 2009, pp. 290–295. 2.1

[24] H. Lund and W. Kempton, “Integration of renewable energy into thetransport and electricity sectors through V2G,” Energy Policy, vol. 36,no. 9, pp. 3578–3587, 2008. 2.1

[25] O. Sundstrom and C. Binding, “Optimization methods to plan the charg-ing of electric vehicle fleets,” in Proceedings of the International Confer-ence on Control, Communication and Power Engineering, 2010, pp. 28–29.2.1, 3.1.2, 3.1.3, B.1, B.1

[26] J. Hu, S. You, J. Østergaard, M. Lind, and Q. Wu, “Optimal charg-ing schedule of an electric vehicle fleet,” in Proceedings of 2011 46thInternational conference on Universities’ Power Engineering Conference(UPEC),. VDE, 2011, pp. 1–6. 2.1, B.1

[27] T. K. Kristoffersen, K. Capion, and P. Meibom, “Optimal charging ofelectric drive vehicles in a market environment,” Applied Energy, 2010.2.1, 3.1.3

[28] T. Lan, J. Hu, Q. Kang, C. Si, L. Wang, and Q. Wu, “Optimal controlof an electric vehicle’s charging schedule under electricity markets,” inNeural Computing and Application, 2012. 2.1, 3.1.2, B.1, B.1, B.1

[29] Y. Cao, S. Tang, C. Li, P. Zhang, Y. Tan, Z. Zhang, and J. Li, “AnOptimized EV Charging Model Considering TOU Price and SOC Curve,”IEEE Transactions on Smart Grid, vol. 3, no. 1, pp. 388–393, 2012. 2.1

68 BIBLIOGRAPHY

[30] Z. Ma, D. Callaway, and I. Hiskens, “Decentralized charging control forlarge populations of plug-in electric vehicles,” in 2010 49th IEEE Confer-ence on Decision and Control (CDC). IEEE, 2010, pp. 206–212. 2.1

[31] Z. Ma, D. S. Callaway, and I. A. Hiskens, “Decentralized Charging Controlof Large Populations of Plug-in Electric Vehicles,” IEEE Transactions onControl Systems Technology, vol. 21, no. 1, pp. 67–78, 2013. 2.1

[32] L. Gan, U. Topcu, and S. Low, “Optimal decentralized protocol for electricvehicle charging,” in Decision and Control and European Control Confer-ence (CDC-ECC), 2011 50th IEEE Conference on, 2011, pp. 5798–5804.2.1

[33] ——, “Optimal decentralized protocol for electric vehicle charging,” PowerSystems, IEEE Transactions on, vol. 28, no. 2, pp. 940–951, 2013. 2.1

[34] A. Ford, “Electric vehicles and the electric utility company,” ENERGYPOLICY, vol. 22, no. 7, pp. 555–570, 1994. 2.1

[35] S. Shao, T. Zhang, M. Pipattanasomporn, and S. Rahman, “Impact ofTOU rates on distribution load shapes in a smart grid with PHEV pene-tration,” 2010 IEEE PES Transmission and Distribution Conference andExposition: Smart Solutions for a Changing World, 2010. 2.1

[36] A. Faruqui, R. Hledik, A. Levy, and A. Madian, “Will Smart Prices InduceSmart Charging of Electric Vehicles?” The Brattle Group Discussion,2011. 2.1, 4.1.1

[37] O. Corradi, H. Ochsenfeld, H. Madsen, and P. Pinson, “Controlling Elec-tricity Consumption by Forecasting its Response to Varying Prices,” IEEETransactions on Power Systems, vol. 28, no. 1, pp. 421–429, 2013. 2.1

[38] R. Yu, W. Yang, and S. Rahardja, “A Statistical Demand-Price ModelWith Its Application in Optimal Real-Time Price,” IEEE Transactionson Smart Grid, vol. 3, no. 4, pp. 1734–1742, 2012. 2.1, 3.2.3

[39] G. T. Heydt, “The Impact of Electric Vehicle Deployment on Load Man-agement Strategies,” IEEE Power Engineering Review, vol. PER-3, no. 5,pp. 41–42, 1983. 2.2

[40] S. Rahman and G. Shrestha, “An investigation into the impact of electricvehicle load on the electric utility distribution system,” Power Delivery,IEEE Transactions on, vol. 8, no. 2, pp. 591–597, 1993. 2.2

[41] K. Clement-Nyns, E. Haesen, and J. Driesen, “The impact of chargingplug-in hybrid electric vehicles on a residential distribution grid,” IEEETransactions on Power Systems, vol. 25, no. 1, pp. 371–380, 2010. 2.2

BIBLIOGRAPHY 69

[42] R. C. Green II, L. Wang, and M. Alam, “The impact of plug-in hybridelectric vehicles on distribution networks: A review and outlook,” Re-newable and Sustainable Energy Reviews, no. 1, pp. 544–553, Jan. 2011.2.2

[43] C. Søndergren, “Edison project: Electric Vehicles in Future Market Mod-els,” Danish Energy Association, Tech. Rep., 2011. 2.2, 2.2, 3.1.1, 3.1.1.2,4.4.1

[44] I. Olle and F. Anders, “Smart charging for electric vehicles,” Energy, pp.33–36, 2012. 2.2, 2.3

[45] J. A. P. Lopes, F. J. Soares, P. M. Almeida, and M. M. da Silva, “SmartCharging Strategies for Electric Vehicles: Enhancing Grid Performanceand Maximizing the Use of Variable Renewable Energy Resources,” inEVS24-The 24th International Battery, Hybrid and Fuel Cell Electric Ve-hicle Symposium and Exhibition, Stavanger, Norway, 2009. 2.2, 2.5

[46] A. Ipakchi, “Demand side and distributed resource management a transac-tive solution,” in Power and Energy Society General Meeting, 2011 IEEE.IEEE, 2011, pp. 1–8. 2.2, 2.6

[47] S. Katipamula, D. P. Chassin, D. D. Hatley, R. G. Pratt, and D. J. Ham-merstrom, “Transactive controls: Market-based gridwisetm controls forbuilding systems,” Pacific Northwest National Laboratory, Richland, WA.,2006. 2.2

[48] D. Hammerstrom, T. Oliver, R. Melton, and R. Ambrosio, “Standard-ization of a hierarchical transactive control system,” Grid Interop, vol. 9,2009. 2.2

[49] M. A. Elizondo, “Usa-argentina collaboration: Smart grid applications indistribution systems,” PNNL, Tech. Rep., 2013. 2.2, 2.7

[50] J. Kok, C. Warmer, and I. Kamphuis, “Powermatcher: multiagent controlin the electricity infrastructure,” in Proceedings of the fourth internationaljoint conference on Autonomous agents and multiagent systems. ACM,2005, pp. 75–82. 2.2, 2.8

[51] H. Akkermans, J. Schreinemakers, and K. Kok, “Microeconomic dis-tributed control: theory and application of multi-agent electronic mar-kets,” in Proceedings of CRIS, 2004, pp. 163–176. 2.2, 4.2.1

[52] R. Lotte Holmberg, B. Christian, and T. Mikael, “Managing congestionin distribution grids-market desing consideration, as part of the readyproject,” NEAS Energy, Tech. Rep., 2012. 2.2, 2.9, 4.1.1, 4.4.1

70 BIBLIOGRAPHY

[53] T. Nagata and H. Sasaki, “A multi-agent approach to power systemrestoration,” IEEE Transactions on Power Systems, vol. 17, no. 2, pp.457–462, 2002. 2.3

[54] C. Rehtanz, Autonomous systems and intelligent agents in power systemcontrol and operation. Springer, 2003. 2.3

[55] D. W. Bunn and F. S. Oliveira, “Agent-based simulation-an applicationto the new electricity trading arrangements of england and wales,” Evo-lutionary Computation, IEEE Transactions on, vol. 5, no. 5, pp. 493–503,2001. 2.3

[56] I. Praca, C. Ramos, Z. Vale, and M. Cordeiro, “Mascem: a multiagent sys-tem that simulates competitive electricity markets,” Intelligent Systems,IEEE, vol. 18, no. 6, pp. 54–60, 2003. 2.3

[57] A. Saleem, “Agent based control of electric power systems with distributedgeneration,” Ph.D. dissertation, Technical University of Denmark, 2010.2.3

[58] M. M. Nordman and M. Lehtonen, “An agent concept for managingelectrical distribution networks,” IEEE Transactions on Power Delivery,vol. 20, no. 2, pp. 696–703, 2005. 2.3

[59] D. Issicaba, M. Rosa, W. Franchin, and J. P. Lopes, “Agent-based systemapplied to smart distribution grid operation.” 2.3

[60] M. Pipattanasomporn, H. Feroze, and S. Rahman, “Multi-agent systems ina distributed smart grid: Design and implementation,” in 2009. PSCE’09.IEEE/PES Power Systems Conference and Exposition. IEEE, 2009, pp.1–8. 2.3

[61] F. Ren, M. Zhang, and D. Sutanto, “A multi-agent solution to distribu-tion system management by considering distributed generators,” IEEETransactions on Power Systems, vol. 28, no. 2, pp. 1442–1451, 2013. 2.3

[62] E. L. Karfopoulos and N. D. Hatziargyriou, “A Multi-Agent System forControlled Charging of a Large Population of Electric Vehicles,” IEEETransactions on Power Systems, vol. 28, no. 2, pp. 1196–1204, 2013. 2.3

[63] J. Miranda, J. Borges, M. J. Mendes, and D. Valerio, “Development ofa multi-agent management system for an intelligent charging network ofelectric vehicles,” in 18th IFAC World Congress, vol. 18, no. 1, 2011, pp.12 267–12 272. 2.3

[64] J. Greunsven, E. Veldman, P. Nguyen, J. Slootweg, and I. Kamphuis, “Ca-pacity management within a multi-agent market-based active distributionnetwork,” in 2012 3rd IEEE PES Innovative Smart Grid Technologies.IEEE, 2012, pp. 1–8. 2.3

BIBLIOGRAPHY 71

[65] S. Rohjans, S. Lehnhoff, S. Schutte, S. Scherfke, and S. Hussain, “mosaik -a modular platform for the evaluation of agent-based smart grid control,”in Innovative Smart Grid Technologies Europe (ISGT EUROPE), 20134th IEEE/PES, 2013, pp. 1–5. 2.3

[66] H. Camilla, T. Mikael, B. Niels Christian, C. Søndergren, and L. H.Hansen, “Introducing electric vehicles into the current electricity mar-kets,” EDISON Consortium, Tech. Rep., 2010. 3.1.1, 3.1.1.2

[67] A. P. Houmoller, “The nordic electricity exchange and the nordic modelfor a liberalised electricity market,” Nord pPool Spot, Denmark, Tech.Rep., 2009. 3.1.1.1

[68] K. M. Rajesh and S. Shashwat, “Market splitting- a mechanism for con-gestion management,” Indian Energy Exchange, 2009. 3.1.1.1

[69] H. Bindner, T. Cronin, P. Lundsager, J. F. Manwell, U. Abdulwahid, andI. Baring-Gould, “Lifetime modelling of lead acid batteries,” Benchmark-ing, vol. 12, p. 82, 2005. 3.1.2

[70] L. Guzzella and A. Sciarretta, Vehicle propulsion systems: introduction tomodeling and optimization. Springer Verlag, 2007. 3.1.2, B.1

[71] S. Shahidinejad, E. Bibeau, and S. Filizadeh, “Statistical Developmentof a Duty Cycle for Plug-in Vehicles in a North American Urban SettingUsing Fleet Information,” IEEE Transactions on Vehicular Technology,vol. 59, no. 8, pp. 3710–3719, 2010. 3.1.3

[72] L. Christensen, “Edison project report: Wp1.3 electric vehicles and thecustomers,” DTU Transport, Tech. Rep., 2011. 3.1.3

[73] Q. Wu, A. H. Nielsen, J. Ø stergaard, S. T. Cha, F. Marra, Y. Chen, andC. Træ holt, “Driving Pattern Analysis for Electric Vehicle (EV) GridIntegration Study,” IEEE Inovative Smart Grid Technology Europe 2010.3.1.3, 4.1.1

[74] R. J. Bessa and M. A. Matos, “Economic and technical management ofan aggregation agent for electric vehicles: a literature survey,” EuropeanTransactions on Electrical Power, no. 3, pp. 334–350, 2012. 3.2.1

[75] T. G. San Roman, I. Momber, M. R. Abbad, and A. Sanchez Miralles,“Regulatory framework and business models for charging plug-in electricvehicles: Infrastructure, agents, and commercial relationships,” EnergyPolicy, vol. 39, no. 10, 2011. 3.2.1

[76] D. Energi and Energinet.dk, “Smart grid in denmark,” Dansk Energi andEnerginet.dk, Tech. Rep., 2011. 4.1.1

72 BIBLIOGRAPHY

[77] S. You and H. Segerberg, “Integration of 100% micro-distributed energyresources in the low voltage distribution network: A danish case study,”Applied Thermal Engineering, no. 0, pp. –, 2013. 4.1.1

[78] K. Qian, C. Zhou, M. Allan, and Y. Yuan, “Modeling of load demand dueto ev battery charging in distribution systems,” Power Systems, IEEETransactions on, vol. 26, no. 2, pp. 802–810, 2011. 4.1.1

[79] D. Wu, D. Aliprantis, and K. Gkritza, “Electric energy and power con-sumption by light-duty plug-in electric vehicles,” Power Systems, IEEETransactions on, vol. 26, no. 2, pp. 738–746, 2011. 4.1.1

[80] A. Ashtari, E. Bibeau, S. Shahidinejad, and T. Molinski, “Pev chargingprofile prediction and analysis based on vehicle usage data,” Smart Grid,IEEE Transactions on, vol. 3, no. 1, pp. 341–350, 2012. 4.1.1

[81] A. Lojowska, D. Kurowicka, G. Papaefthymiou, and L. Van der Sluis,“Stochastic modeling of power demand due to evs using copula,” PowerSystems, IEEE Transactions on, vol. 27, no. 4, pp. 1960–1968, 2012. 4.1.1

[82] J. Northcote-Green and R. G. Wilson, Control and automation of electricalpower distribution systems. CRC Press, 2006. 4.1.2

[83] I. distribution automation working group, “Smart grid for distributionsystems: the benefits and challenges of distribution automation (da) (draftversion 2) white paper for nist,” IEEE PES Society, Tech. Rep., 2013. 4.1.2

[84] ALSTOM, “Alstom grid smart distribution solutions for smart utilitieswhite paper,” ALSTOM, Tech. Rep., 2013. 4.1.2

[85] S. H. Clearwater, Market based control: a paradigm for distributed resourceallocation. World Scientific, 1996. 4.2.1

[86] M. P. Wellman, “A market-oriented programming environment and itsapplication to distributed multicommodity flow problems,” Journal of Ar-tificial Intelligence Research, vol. 1, no. 1, pp. 1–23, 1993. 4.2.1, 4.2.2

[87] J. Q. Cheng and M. P. Wellman, “The walras algorithm: A convergentdistributed implementation of general equilibrium outcomes,” Computa-tional Economics, vol. 12, no. 1, pp. 1–24, 1998. 4.2.1, 4.2.2

[88] N. C. Nordentoft, “ipower wp3.8 development of a dso-market on flexibil-ity services,” iPower Consortium, Tech. Rep., 2013. 4.2.1

[89] M. Grant and S. Boyd, “CVX: Matlab software for disciplined convexprogramming, version 2.0 beta,” http://cvxr.com/cvx, Sep. 2013. 4.2.1

[90] B. Moradzadeh and K. Tomsovic, “Two-stage residential energy manage-ment considering network operational constraints.” 4.2.1, 4.4.2

http://cvxr.com/cvx

BIBLIOGRAPHY 73

[91] T. W. Sandholm, “Distributed rational decision making,” Multiagent sys-tems: a modern approach to distributed artificial intelligence, pp. 201–258,1999. 4.2.2

[92] R. Sebastian, R. Christian, and L. Sebastian, “On the problem of control-ling shiftable prosumer devices with price signals,” in submitted to 18thPower Systems Computation Conference (PSCC), 2014. 4.4.2

[93] N. Gatsis and G. Giannakis, “Residential load control: Distributedscheduling and convergence with lost ami messages,” Smart Grid, IEEETransactions on, vol. 3, no. 2, pp. 770–786, 2012. 4.4.2

[94] N. R. Jennings and S. Bussmann, “Agent-based control systems,” IEEEcontrol systems, vol. 23, no. 3, pp. 61–74, 2003. 5.1, 5.1

[95] T. Sarosh, d. Pedro, and M. Seshashayee, “organizations for computer-based agents,” Engineering Intelligent Systems, vol. 2, pp. 75–87, 1993.5.1

[96] N. Howden, R. Ronnquist, A. Hodgson, and A. Lucas, “Jack intelligentagents-summary of an agent infrastructure,” in 5th International confer-ence on autonomous agents, 2001. 5.2.2

[97] B. Mads, S. N. Michael, and B. R. Theis, “The flexibility clearinghouse, abidding framework,” Technical University of Denmark, Tech. Rep., 2013.5.2.2

[98] H. Hansen, H.-H. Holm-Hansen, O. Samuelsson, L. H. Hansen, H. W.Bindner, H. Johannsson, and P. Cajar, “Coordination of system needsand provision of services,” in Electricity Distribution (CIRED 2013), 22ndInternational Conference and Exhibition on, 2013, pp. 1–4. 1

[99] S. You, J. Hu, K. Heussen, and C. Zhang, “Analytical framework formarket-oriented dsr flexibility integration and management,” Energy andPower Engineering, vol. 5, p. 1367, 2013. 6.2

[100] J. Hu, S. You, M. Lind, and J. Østergaard, “Coordinated charging ofelectric vehicle for congestion prevention in the distribution grid,” acceptedfor the publication on IEEE Transactions on smart grids, 2013. B.1

[101] X. Han, “Quantitaive analysis of distribtued energy resources in future dis-tribution networks,” Master’s thesis, Royal Institute of Technology, 2012.B.2

[102] R. D. Christie, B. F. Wollenberg, and I. Wangensteen, “Transmissionmanagement in the deregulated environment,” Proceedings of the IEEE,vol. 88, no. 2, pp. 170–195, 2000. B.3

74 BIBLIOGRAPHY

[103] R. G. Smith, “The contract net protocol: High-level communication andcontrol in a distributed problem solver,” Computers, IEEE Transactionson, vol. C-29, no. 12, pp. 1104–1113, 1980. B.4

[104] R. D. Zimmerman, “Uniform price auctions and optimal power flow,”Matpower Technical Note 1, February 2010., Tech. Rep., 2010. B.4

[105] S. Boyd, L. Xiao, and A. Mutapcic, “Subgradient methods,” StanfordUniversity, Autumn Quarter, vol. 2004, 2003. B.5, B.5

[106] S. Boyd, L. Xiao, A. Mutapcic, and J. Mattingley, “Notes on decomposi-tion methods,” Stanford University, 2007. B.5, B.5

[107] J. McCalley, Z. Zhong, V. Vishwanathan, and V. Honavar, “Multiagentnegotiation models for power system applications,” Autonomous systemsand intelligent agents in power system control and operation, C. Rehtanz,editor, Springer-Verlag, Berlin, pp. 49–74, 2003. B.6

[108] H. Raiffa, The art and science of negotiation. Harvard University Press,1982. B.6

[109] M. Kraning, E. Chu, J. Lavaei, and S. Boyd, “Dynamic network energymanagement via proximal message passing,” Optimization, vol. 1, no. 2,pp. 1–54, 2013. B.7

Appendix A

Core publications

A.1 Optimization and control methods for smartcharging of electric vehicles facilitated byfleet operator: review and classification

This paper is published in International Journal of Distributed Energy Resourcesand Smart grids, Volume 10, Number 1, 2014.

OPTIMIZATION AND CONTROL METHODS FOR SMART CHARGING OF ELECTRIC VEHICLES

FACILITATED BY FLEET OPERATOR: REVIEW AND CLASSIFICATION

Junjie Hu1, Shi You1, Chengyong Si2, Morten Lind1, Jacob Østergaard1 1Department of Electrical Engineering, Technical University of Denmark

Building 325, Elektrovej, 2800, Lyngby, Denmark. E-mail: [email protected]

2College of Electronics and Information Engineering, TongJi University, Shanghai, China

Keywords: Electric vehicles; Smart charging; Review and classification; Direct control and price control

ABSTRACT

Electric vehicles (EV) can become integral parts of a smart grid, since they are capable of providing valuable services to power systems other than just consuming power. As an important solution to balance the intermittent renewable energy re-sources, such as wind power and PVs, EVs can absorb the energy during the period of high electricity penetration and feed the electricity back into the grid when the demand is high or in situations of insufficient electricity generation. However, the extra loads created by increasing EVs may have adverse impacts on grid. These factors will bring new challenges to the utility system operator; accordingly, smart charging of EVs is needed. This paper presents a review and classification of methods for smart charging of EVs found in the literature. The study is mainly executed from the control theory perspectives. Firstly, service dependent aggrega-tion and the facilitator EV fleet operator are introduced. Secondly, control architec-tures and their integrations in term of electricity market and distribution grid are discussed. Then, data analysis of EVs including a battery model and driving pattern is presented. Further discussion is given on mathematical modelling and control of smart charging of EVs. Finally, the paper discusses and proposes future research directions in the area.

1 INTRODUCTION EVs are commonly recognized as smart grid assets in addition to their primary transport function. They can be utilized to balance power fluctuations caused by the high penetration of intermittent renewable energy sources [1], [2]. However, a large scale application of EVs also mean new loads to electric utilities, and unde-sirable peaks may exist in the distribution network when recharging the battery [2]. All these factors bring new challenges to the system operator. As a result, smart charging (including power to vehicle and vehicle to grid (V2G)) solutions are needed which can make EV an asset to the grid rather than a mere traditional load and make the grid more flexible.

Much research has been done to address the above challenges. The purpose of this study is to give a review and classification of the control strategies used for smart charging of EV fleets. From the literature, it is summarized and concluded that a new business entity, namely the EV fleet operator (FO) has been widely proposed capturing the new business opportunities by providing the multiple services of EVs and then by this contributing to the challenges solving of power distribution system operator. Alternatively names for an EV FO are used such as EV virtual power plant, EV aggregator, EV charging service provider or EV service provider (EVSP). The new entities [3], [4] could be independent or integrated in an existing business function of the energy supplier or distribution system operator.

In principle, two types of control architectures are used by FOs when aiming at the above objectives, named centralized and decentralized control. Centralized control means electric vehicles can be aggregated and controlled by FO directly, while the decentralized control usually is implemented in the form of price signal, i.e. the individual EV optimizes the charging based on the electricity price information made available to them either from EV FO or the utility. A comprehensive discus-sion and comparison on these architectures can be found in [5], [6]. From the dis-cussions in [5]-[7], it can be shortly summarized that for a centralized charging the decisions are made on the system-level and therefore can give better results such as ensuring the safety of the distribution network; however, the cost of communica-tion infrastructure would be high for centralized charging. For a decentralized charging, one of main advantages is the possibility to minimize the communica-tions infrastructure cost [8], however, the solution may or may not be optimal, de-pending on the information sharing and methods used to make the charging scheme.

The paper is organized as follows: The control objectives are discussed in Section 2. Section 3, 4 describes the role and control architectures of EV FO. The battery model and driving patterns of EVs are briefly discussed in section 5. Some com-monly used algorithms in the centralized and decentralized control of smart charg-ing of EVs are presented in Section 6 and 7, respectively. Section 8 concludes the paper with some suggestions for future research.

2 SERVICE DEPENDENT AGGREGATION In [9], Lopes et al. shortly summarized that a large deployment of EVs will involve the following studies: 1) Evaluation of the impacts that battery charging may have on system operation; 2) Identification of adequate operational management and control strategies regarding batteries’ charging periods; 3) Identification of the best strategies to be adopted in order to use preferentially renewable energy sources (RES) to charge EVs; 4) Assessment of the EV potential to participate in the provi-sion of power system services, including reserves provision and power delivery, within a vehicle to grid (V2G) concept. Inspired by this summary, we will first review four kinds of goals when investigating in smart charging of an EV fleet. In addition, we also see these four objectives as four types of opportunities and prod-ucts that can be captured by FOs and then provided to other actors in a smart grid context.

2.1 Providing ancillary services to the transmission system operator (TSO) Kempton et al. [10], [11] analysed the potential profits of V2G support by compar-ing it to existing ancillary services and found that participating regulation power market appears to be most promising and offers a substantial earning potential to EV owners. Rotering and Ilic [12] took into account vehicle to grid as a mean of generating additional profits by participating in the ancillary service markets. Based on the data of the independent system operator of California, provision of regulating power substantially improves plug-in hybrid electric vehicle economics and the daily profits amount to $ 1.71, including the cost of driving. Han et al. [13] proposed an FO that manages EVs to provide frequency regulation services, the cost arising from the battery charging and the revenue obtained during the partici-pation is investigated. The problem is formulated as an optimization problem and dynamic programming is used to generate the charging control profile. Divya et al. [14] carried out a study investigating the feasibility of integrating EVs in the Dan-ish electricity network which is characterized by high wind power penetration. They found that EVs have the potential to assist in integrating more wind power in 2025 when the EV penetration levels would be significant enough to have an im-pact on the power systems. Tuffner and Meyer [15] explored two different charg-ing schemes: V2G Half and V2G Full to handle the entire additional energy imbal-ance imposed by adding 10GW of additional wind to the Northwest Power Pool. The result indicates that the proposed frequency based charging strategy can meet the new balancing requirements. However, this also depends on the charging sta-tion availability (residential and public charging station), the economics of the im-plementation and a viable and compelling business model. All these results indicate that it is reasonable and profitable to participate in the electricity market and pro-vide ancillary service to the gird.

2.2 Providing services to renewable energy source (RES) supplier Lopes et al. [16] investigated the dynamic behaviour of an isolated distribution grid when wind power and electric vehicles are presented. The objective is to quantify

the amount of intermittent RES that can be safely integrated into the electric power system with the utilization of EVs’ storage capacity. Another study [17] by the same author analyse two tasks. The first part of the work studied the maximum share of EVs on the low voltage networks without violating the system’s technical restrictions. The second part focused on the prevention of wasting renewable ener-gy surplus when charging the EVs. The results indicate that the grid can allocate higher penetration of EVs with a smart charging strategy compared with a dumb charging and that the EVs have the capability to store energy and discharge to grid later into the system. In this way, the RES can be utilized more. Lund and Kempton [18] investigated the impact of using V2G technology to integrate the sustainable energy system. Two national energy systems are modelled; one for Denmark in-cluding combined heat and power (CHP), the other is a similarly sized country without CHP. The model (EnergyPLAN) integrates energy for electricity, transport and heat, includes hourly fluctuations in human needs and the environment (wind resource and weather-driven need for heat) The results indicated that adding EVs and V2G to these national energy systems allows integration of much higher levels of wind electricity without excess electric production, and also greatly reduces national CO2 emissions.

2.3 Minimizing charging cost An electricity market is presumed and is ideally suited for the application of opti-mal charging control; this is because the various hourly market prices can bring benefits for EVs if they are scheduled to charge in the period of lower prices. With-in this scope, most the work [19]-[21] assume that the EV FO manages the electric-ity market participation of an EV fleet and presents a framework for optimal charg-ing or discharging of the EVs. In addition, the electricity price of the day-ahead spot market and the regulation market and the driving patterns of the EV fleet are usually assumed to be known by the FO who is assumed to be the price-taker in the electricity market in studies. However, Kristoffersen et al. [22] also investigated the possibilities of EV management where the FO has a significant market share and can affect electricity prices by changing the load through charging and dis-charging. Besides studying the optimal charging from an EV fleet perspective, research in [12], [23] showed how dynamic programming can be utilized by the individual EV controller to make an optimal charging schedule taking into account the electricity market price. In [24], an intelligent charging method is proposed which responds to TOU price and minimize the charging cost.

2.4 Providing ancillary services to distribution system operator (DSO) It is assumed that the distribution network has the capacity to allocate new loads when achieving the objectives discussed above. With the objective of avoiding grid bottlenecks, the purpose of the smart charging is to solve the potential grid conges-tion problem. Many investigations has been performed studying the impact of EVs on grid, which can be dated back to the early 1980s [25]. In [26], the authors gave a review and outlook about the impact of EVs on distribution networks. Sundstrom

and Binding [27] considered the power grid on the Danish island of Bornholm, where the grid of the isolated island is used to study the impact of EVs and the potential profit to be made of grid services. The focus of the paper is on proposing a method for planning the individual charging schedules of a large EV fleet as well as respecting the constraints in the low-voltage distribution grid. The impact of EVs on the electricity grid is studied in [28], where the focus is on the Vermont power grid. They assume a dual-tariff, nightly charging scheme, and conclude that enough transport capacity is available in the power grid. Lopes et al. [29] studied the potential impact on a low-voltage distribution grid. Smart charging behaviour is here considered to maximize the density of EV deployment into the grid, i.e., to reach the maximally tolerable number of EVs and meanwhile maintaining grid constraints. Kristien et al. [30] investigated the impact of charging EVs on a resi-dential distribution grid and illustrated the results of coordinated and uncoordinated charging. Without coordination of the charging, the power consumption on a local scale can lead to grid problems. While the coordination of the charging can reduce the power losses, power quality is improved to a level which is similar to the case where no EVs are present.

2.5 Analysis of the research framework and the goals of smart charging Several questions would naturally arise after reviewing the four goals described in 2.1 to 2.4, e.g., whether some goals can be integrated when making the optimal charging schedules of an EV fleet, what are the relationships between these four goals. In [12], the authors took into account vehicle to grid as a mean of generating additional profits by participating in the ancillary service markets and integrated it with the goal of minimizing the charging cost of the EV. The result indicated that the combined goals substantially improve EV economics. Sundstorm and Binding [27] considered the distribution grid congestion issue when minimizing the charg-ing cost of an EV fleet. It is observed that multi-goals study is already performed, however, a systematic way of understanding the relationships between the de-scribed goals is missing.

In general, relationships between goals can be described as [31]:

• Independence: the goals do not affect each other. • Cooperation: achieving one goal makes it easier to achieve the other. • Competition: one goal can be achieved only at the expense of the other. • Interference/Coordination: one goal must be achieved in a way that takes

the other goal into account.

We use these four relationships as guideline and analyse the relationships between the four goals of smart charging. Table 1 presents the results.

Table 1: Relationships between the four goals discussed above

Providing services to RES supplier

Providing ancillary ser-vices to TSO

Minimizing charging cost

Providing ancillary ser-vices to DSO

Providing ser-vices to RES supplier

N.A. Cooperation Cooperation Coordination

Providing ancil-lary services to TSO

Cooperation N.A. Cooperation Coordination

Minimizing charging cost Cooperation Cooperation N.A. Coordination

Providing ancil-lary services to DSO

Coordination Coordination Coordination N.A.

It is shown in Table 1 that the first three goals needed to be coordinated with the last one, and this coordination is usually called congestion management in distribu-tion network and the topic has recently attracted many researches. Besides, Table 1 shows that the first three objectives can be well integrated when generating the optimal schedule of EV fleets. With this qualitative analysis, it is beneficial for the FO to make global optimal schedules.

3 INTRODUCTION OF FO IN THE CONTEXT OF SMART GRIDS From previous discussion, despite some services like minimizing charging cost could be done by individual EV, in most cases, these services can be practical in place only provided by a large fleet of EV. As shortly mentioned above, FO is widely proposed to aggregate the large penetration of EVs in the near future (FO used in the Edison project: http://www.edison-net.dk/). Firstly, the roles of the FOs are summarized from the literature; then we show the relationships between the FOs and other actors in a smart grid context; further discussion is made on the communication standard used for implementing the charging schedules.

3.1 Role of FOs Tomas et al. [4] proposed two new electricity market agents: the EV charging man-ager and the EV aggregator/FO which are in charge of developing charging infra-structure and providing charging services, respectively; based on this, the authors proposed a regulatory framework for charging EVs. Similar concept is introduced in [32], where the concept of EV service provider (EVSP) is discussed. In [32], the EVSP has two functions: one is responsible for installing and operating the charg-

ing equipment, another is supplying electricity to the EVs. In term of the feasibility of applying the FO concept, Bessa and Matos [3] gave a literature review regarding the economic and technical management of an aggregation agent for electric vehi-cles. The reviewed papers are organized into three technical categories: electricity market and EV technical and economic issues; aggregation agent concept, role and business model; algorithms for EV management as a load/resource.

It is observed that the main difference between the proposed solutions of FO lies on whether the FO has twofold functions or sole function, i.e., some studies assumed that a FO functions as both charging equipment supplier and charging service pro-vider, others only refer FO as the charging service provider. Although various dif-ferences exist in the details of the proposed FO concepts, they are assumed to achieve the same goals in this study, regardless the ownership of the charging equipment:

• Guarantee driving needs of the EV owners with optimal management of EV charging;

• Provide ancillary services to power system operators with optimal alloca-tion of EV fleet resources.

3.2 Service relationships between FOs and other actors in a smart grid Fig.1 illustrates the relationships between FOs and other actors in a smart grid by showing the services that FOs can provide to them.

FOsFleet Operation

TSO

Grid Measurements

DSO

Conventional load

Information flow

Physical connection

Providing ancillary service

Providing ancillary service

Minimizing charging cost

Enabling renewable energy

EV ownerLocal electricity producer

Fig.1: The services relationships between FO and other actors in a smart grid.

Note that the relationship between FOs and EVs is a slightly more complex. From one perspective, FOs need to attract the participation of EVs and then have the ability to provide services to other actors in the smart grid; from another perspec-

tive, FOs can provide the service of minimizing charging cost to EVs which help the EV owner to save money. Therefore, FO may need to consider many factors rather than purely make benefits when providing services to EVs.

3.3 Implementing the charging services/schedule provided by FOs With the purpose of illustrating how the charging schedule is implemented, this section discusses the relevant communication standard for integrating EVs into the distribution grid. For example, the studies in [12], [13], [19], [22] focus on generat-ing the optimal charging schedule instead of implementing it. These parts are sup-plemented by the works in [33]-[37]. It is noted that the purpose is to provide the relevant/widely used communication standard which can support the EV smart charging rather than comparing the various communication standards. Su et al. [33] presented a overview of EVs from the perspectives: 1) charging infrastructure (so-ciety of automotive engineers standard) and Plug-in Hybrid Electric Vehicles (PHEVs)/Plug-In Electric Vehicles (PEVs) batteries, 2) communication require-ments. In European area, studies in [34]-[37] recommended the IEC standards which are illustrated in Fig.2. The objectives of all the studies [34]-[37] are the realization of a standardized communication interface, the vehicle to grid commu-nication interface. The standardization will make it possible for users of EVs to have easy access to EV charging equipment (EVSE) and related service throughout Europe. EVSE refers to all the devices installed for the delivery of power from the electrical supply point to the EV. EVSE supports the smart charging functions. The decision can be made on the EV level or on the FOs level. The IEC 15118 is the most recommended communication standard in the work [34]-[37] and demon-strated in details in [34], [35] by showing the sequence diagram of a charging pro-cess between the EVSE and the EVs. For the communication between the EVSE and the FOs, it is recommended that IEC 61850 can fulfil the functions.

FOs

Public network

EV owner

EVSE

IEC 61850IEC 15118

EVi

Figure 2: Relevant ICT standards support the EV smart charging in the

context of smart grids

In general, we define EVi as the combination of the EVSE and the EV as well as holding the intelligence endowed by the EV owner (illustrated in the Fig.2). With

this background, for the next parts of this paper, we will review the control archi-tectures, the algorithms which are used by FOs in the literature and the communi-cation part will be ignored.

4 CONTROL ARCHITECTURES AND THEIR INTEGRATION

4.1 Centralized charging control of EV FO

FO operation

• Modeling the state of charging of the battery with linear or nonlinear approximation

• life time assessing model, battery degration analysis

• Day-ahead market• Regulation power market• Other reserve market

• Forecasting driving requirement of EV owner

• Location prediction of Evs• Normal vehicle’s historical

data are used• Normal distribution,

Monte-Carlo based method

• Thermal constraint of cable and transformers

• Voltage constraint

Battery Model

EV FO

Electricity Market

Driving Pattern

Grid Contraint

EV1 EV2 EVNEViControl Signal

Information flow

Figure 3: Primary inputs and output of EV FO.

Fig. 3 mainly depicts the four inputs when making the control strategies. In this context, FO obtains all the relevant information including the EV battery model, the EV driving patterns, the grid constraint and the electricity price and centrally makes the charging schedule for each EV. In contrast, some EV owners want to generate the charging schedule by themselves, this is called decentralized control. However in the context of decentralized charging, the FO still needs to coordinate the grid constraints with the EV owners and this coordination is usually imple-mented by using price signal. In the following section, we will present two meth-ods of implementing decentralized charging control as well as respecting to the grid constraint.

4.2 Decentralized charging management of EV FO The scheme of the information flow in decentralized charging control is presented in Fig.4.

Charging profile of vehicle 1,...N

Price information

EV FO perspective• Send the updated price/

aggregated schedule

EV owner perspective• Intelligent Individual controller• sophisticated control equipment• Send back the charging profile

EV FO

EV1 EVNEVi

Price information

EV FO

EV1 EVNEVi

EV FO perspective • Predict the user’s response/price

elasticity• Determine the optimal price signal

EV owner perspective• Programmable appliance timers• The price flexible consumer

• One way price signal /communication• Two way Price signal/communication

Figure 4: Schematic view of the information flow between the FO and the

EVs.

In Fig.4, two kinds of price signals are presented. For the left figure, i.e., two way price signal, this is also used such as game theory, valley filling. The basic idea is that EVs update their charging profiles independently given the price signal; the FO guides their updates by altering the price signal. Several iterations are required for the implementation. For the right figure, we call it one way price signal method; this method requires FO to predict the users’ response to the prices. The price sig-nal can be designed simply as time-of-use price or more dynamic prices.

4.3 Comparison between control strategies Table 2 compares the two control methods based on literature review. We first clarify the terminologies used in this study: centralized control and decentralized control are regarded as architectures, which mean the charging schedule decision is made either in upper FO level or local EV controller level. Direct control means that FO sends the control signal to the EVs and the EVs executes the charging schedule. Price control means that the FO coordinate their requirements (distribu-tion grid constraints) by sending electricity price to the EV controller and the EV controller takes the decision to generate the charging schedule. This is indirect control for the FO because the FO is only specifying a constraint (the price) for the charging schedule and not the charging schedule itself.

Table 2: Comparison between direct control and price control strategy

Control Methods

Direct control Price control

Features of the control method

• Control signals (i.e., set points)

• High level controller makes the decision

• Price incentive • Consumer make

decision

Advantages • High certainty • Better optimal results

• Privacy improved • Less communica-

tion cost

Disadvantages

• Inflexible • High communication

cost • High computation re-

quirement

• Lower certainty • Better knowledge

on customer’s re-sponse to price re-quired

5 DATA ANALYSIS OF BATTERY MODEL, DRIVING PATTERN

5.1 Battery model Basically, there are two ways to model the charging characteristics of the EV, i.e., the battery model. One is an individual battery pack model, another is aggregated (characterize the state of charge of an EV fleet in one model). For simplicity, most of the studies considered EV as a battery pack when investigating the optimal charging and discharging problem. Currently, most battery model studies [38]-[40] focus on three different characteristics:

• The first and most commonly used model is termed as a performance or a charge model and focuses on modelling the state of charge of the battery, which is the single most important quantity in system assessments.

• The second type of model is the voltage model, which is employed to model the terminal voltage so that it can be used in more detailed model-ling of the battery management system and the more detailed calculation of the losses in the battery.

• The third type of model is the lifetime model used for assessing the impact of a particular operating scheme on the expected lifetime of the battery.

We give further discussion on the first model, usually, linear and nonlinear approx-imation are used to characterize the state of charge of the battery. Linear approxi-mation are utilized in works [19]-[21] to approach the charging behaviour of an EV battery. Rotering and Ilic [12] considered a nonlinear battery model. The studies [20], [21] has shown that violations of the battery boundaries when applying the charging schedule based on the linear approximation are relatively small, i.e., less

than 2% of the usable capacity. The benefit of using a nonlinear approximation does not justify the increase in computation time.

5.2 Driving pattern The analysis of driving pattern can be divided into two main directions:

• Utilization of EVs, in other words, a typical user daily life means that at any point during the day an EV could possibly be in the garage, in an em-ployer’s parking lot, in a store parking lot or on the road. This means that the aggregator needs to characterize/predict the driving pattern of EVs.

• Location of EVs when charging and how many of them will be charged at a time, since such driving patterns produce an impact on the distribution gird.

In most papers [19]-[21] the authors assume that the aggregator know the users’ driving patterns. There are few studies on investigating the driving pattern issue. Kristoffersen et al. [22] investigated the method to construct driving patterns with the historic data in Danish case. By clustering the survey data on the vehicle fleet in Western Denmark (January 2006-December 2007), a representative driving pat-terns for each vehicle user are constructed. S. Shahidinejad et al. [41] developed a daily duty cycle which provides a complete data set for optimization of energy requirements of users and furthermore, this information can also be used to analyse the impact of daytime charging by a fleet of plug-in electric vehicles on the electric utility grid that may create a peak demand during the day to be met by the local utility grid. Normally, intra city or short term driving patterns are largely predicta-ble due to fixed working hours and fixed business schedules and routes.

6 MATHEMATICAL MODELING AND CONTROL: CENTRALIZED CONTROL

In this section, we will present algorithms often used for the centralized control. Linear programming, quadratic programming, dynamic programming and stochas-tic programming will be shown for the discussion through an extensive literature review. Further, a qualitative comparison among the four algorithms will be pre-sented in the end of this section.

6.1 Linear programming (LP) Sundstrom and Binding [20], [21] used linear approximation to characterize the state of charge of a battery and formulate the charging process for an EV fleet into a linear programming based optimization problem:

bT

s Pctmin (1) Subject to

≤≤≥≥≥

ubl

bbb

gbg

sbs

bPbbPAbPAbPA

(2)

With the time slot st , cost vector 𝑐, the charging power 𝑃𝑏, the stopover inequality constraints (As, bs), the generation inequality constraints (Ag, bg), the battery ine-quality constraints (Ab, bb), and the upper and lower bounds (bu, bl). The solution of this linear optimization problem is the optimal charging profile while minimizing the charging cost of EV fleet.

6.2 Quadratic programming (QP) A nonlinear approximation (quadratic formulation) of the battery charging model is also studied in [20], [21]. The results showed that the number of constraints is higher and increases faster with a growing fleet in the quadratic formulation than in the linear formulation, the difference in calculation time increase with increasing fleet size. An example is conducted for comparison and the result indicated that calculating time using the quadratic formulation is 819 times the calculation time using the linear formulation. But the result difference does not justify the benefits of using quadratic formulation. Another example of using quadratic programming method was introduced by Kristien et al. [30] who formulated the power loss prob-lem caused by large penetration of EVs in the grid into a sequential quadratic opti-mization problem. The objective is to minimize the power losses which are treated as a reformulation of the nonlinear power flow equations. The charging power obtained by the quadratic programming cannot be larger than the maximum power of the charger 𝑃𝑚𝑎𝑥. The batteries must be fully charged at the end of cycle, so the energy which flows to the batteries must equal the capacity of the batteries 𝐶𝑚𝑎𝑥. 𝑥𝑛 is zero if there is no EV connected and is one if there is an EV connected at node n. The above problem specification can be represented as follow:

min � � 𝑅𝑙 . 𝐼𝑙,𝑡2𝑙𝑖𝑛𝑒𝑠

𝑙=1

𝑡𝑚𝑎𝑥

𝑡=1

Subject to

�∀𝑡,∀𝑛 ∈ {𝑛𝑜𝑑𝑒𝑠}: 0 ≤ 𝑃𝑛,𝑡 ≤ 𝑃𝑚𝑎𝑥

∀𝑛 ∈ {𝑛𝑜𝑑𝑒𝑠}:∑ 𝑃𝑛,𝑡 .∆𝑡. 𝑥𝑛 = 𝐶𝑚𝑎𝑥𝑡𝑚𝑎𝑥𝑡=1𝑥𝑛 ∈ 0,1

(3)

The quadratic programming techniques are applied using both deterministic and stochastic methods in Kristien’s paper. The input variables in both cases are the daily/hourly load profile. In the deterministic case, the load profiles are static. In

the stochastic case, the load profile are transformed into probability density func-tions, which means that the fixed input parameters are converted into random input variables with normal distributions assumed at each node. The details of stochastic case are presented in the following section.

6.3 Dynamic programming (DP) Dynamic programming is widely used in many papers [12], [13], [23], [30] with different purposes. We introduce the work in [12]. In the paper, a specific control strategy is denoted by

𝜋 = {𝑢0,𝑢1, …𝑢𝑘 , … ,𝑢𝑁−1}

Where 𝑢𝑘 is the control variable denotes a dimensionless and discrete representa-tion of 𝑃𝑘. 𝑃𝑘 corresponds to the purchased power flow. The total cost of a whole charging sequence, 𝐽𝜋 is then given as below:

𝐽0𝜋(𝑥0) = 𝐽𝑁(𝑥𝑁) + � 𝑙𝑘(𝑥𝑘,𝑢𝑘 ,𝑘)𝑁−1

𝑘=1

𝐽𝑁 means cost of the final step, 𝑙𝑘(. ) denotes the cost-to-go for all other steps, N denotes the total number of time intervals. The objective is to find the optimal con-trol variables which can minimize the total cost. The detailed mathematic formula of cost of final step and cost-to-go are not presented here. The purpose of the func-tion used for calculation of cost of final step is ensuring that the battery is fully recharged before the first trip of the following morning. For the function of cost-to-go, the electricity price, regulating-up price and regulating-down price are consid-ered.

This is a classical dynamic programming formulation and the optimal trajectory is calculated starting with the cost of the last state and going backwards through time until the first state’s optimal cost 𝐽00(𝑥0) is given by the algorithm. Concerning the computing time of dynamic programming, the results in [30] show that the differ-ence of the charging profiles for the QP and DP technique are negligible, however, considering the computational time and storage requirements, the storage require-ments are heavier for the DP technique compared to the QP technique, hence, the computational time for DP technique is longer.

6.4 Stochastic programming Most of the current researches [12], [19], [21] assume that the load profiles, initial state of charge, driving pattern, grid conditions and electricity price are known and determined to the FO, however, this is certainly not the case in the reality. It is therefore necessary to put efforts on stochastic approach to reduce the risks, and some works have been done recently [30], [42]-[45].

A stochastic approach for calculation of the daily load profiles is considered in [30] when minimizing the power loss problem. A sample average approximation meth-

od [46] is utilized to formulate the random inputs and the lower bound estimate principle is used to estimate the optimal value. It is noted that the model is the same as presented in equation (3) of this section (section 6.2). The uncertainties of these parameters can be described in terms of probability density functions. In that way, the fixed input parameters are converted into random input variables with normal distributions assumed at each node. N independent samples of the random input variable 𝜔𝑗, the daily load profile, are selected. Following equation (4) gives the estimation for the stochastic optimum 𝑣�𝑛. The function 𝑔(𝑃𝑛,𝑡,𝜔𝑗) gives the power losses and 𝑃𝑛,𝑡 is the power rate of the charger for all the EVs and time steps. 𝑓𝑁 is a sample-average approximation of the objective of the stochastic pro-gramming problem:

𝑣�𝑛 = 𝑚𝑖𝑛{𝑓𝑁(𝑃𝑛,𝑡) ≡ 1𝑁∑ 𝑔�𝑃𝑛,𝑡,𝜔𝑗�𝑁𝑗=1 } (4)

The mean value of the power losses, 𝐸(𝑣�𝑛), is a lower bound for the real optimal value of the stochastic programming problem, 𝑣∗, as shown in the below:

𝐸(𝑣�𝑛) ≤ 𝑣∗

𝐸(𝑣�𝑛) can be estimated by generating M independent samples 𝜔𝑖,𝑗 of the random input variable each of size N. M optimization runs are performed in which the non-linear power flow equations are solved by using the backward-forward sweep method. The optimal values of M samples constitute a normal distribution:

𝑣�𝑛𝑗 = 𝑚𝑖𝑛 �𝑓𝑁�𝑃𝑛,𝑡� ≔

1𝑁�𝑔(𝑃𝑛,𝑡 ,𝜔𝑖,𝑗)𝑁

𝑖=1

� , 𝑗 = 1, … ,𝑀.

𝑣�𝑛𝑗 is the mean optimal value of the problem for each of the M samples. 𝐿𝑁,𝑀 is an

unbiased estimator of 𝐸(𝑣�𝑛). Simulations indicate that in this type of problem, the lower bound converges to the real optimal value when N is sufficiently high:

𝐿𝑁,𝑀 = 1𝑀∑ 𝑣�𝑁

𝑗𝑀𝑗=1 .

A forecasting model for the daily load file for the next 24 hours is required at first, then the daily profile of the available set are varied by a normal distribution func-tion. The standard deviation 𝜎 is determined in such a way that 99.7% of the sam-ples vary at maximum 5% or 25% of the average. In general, the simulation results indicated that the difference between the power losses of the stochastic and the deterministic optimum is rather small.

Other studies such as Fluhr et al. [42] use Monte-Carlo method to generate the probability distributions of the driven travel paths for one week with the survey ”Mobility in Germany” (MIG), because the original data MIG only provide one day driving behavior; studies in [43]-[45] use normal distribution and Poisson distribution to investigate the probabilistic distribution of plugin time and initial state of charge of EVs.

6.5 A summary of the presented algorithms with three types of criteria In table 3, we mainly summarize the information of the presented algorithms in term of the computation time, the certainty of performance, and the applicability. The summary aggregates the comparisons described in the literatures in term of computation time and performance of the presented algorithms. Besides, the ap-plicability of the presented algorithms is summarized from two perspectives.

Table 3: General comparison between the presented algorithms

Computation time

Certainty of Performance

Applicability in general

Applications to EV charg-ing

Linear

programming Used in: [19], [20], [21].

Generally, it is the fastest.

Results in [19], [20], [21] showed that the perfor-mance is excellent in term of finding the optimal solu-tion.

1) The objective function is linear, and the set of constraints is specified using only linear equali-ties and inequalities.

2) Standard model, easy for implementation.

Minimize charg-ing cost of EVs.

Quadratic programming Used in: [20], [21], [30].

Ref. [20] showed that the calcula-tion time using the QP is 819 times than the one using LP for a fleet of 50 vehi-cles.

Ref. [20] showed that the difference between using LP and QP is minor. Therefore, the benefit of using the QP does not justify the increase in computation time.

1) The objective function has quadratic terms, while the feasible set must be specified with linear equalities and inequalities.

2) Standard model, easy for implementation.

1) Minimize charging cost of EVs.

2) Minimize power losses of power systems.

Dynamic

programming Used in: [12], [13], [23], [30].

Ref. [30] indicat-ed that the com-putational time for DP is slower compared to QP.

Ref. [30] showed that the difference between the charg-ing profile of using QP and DP is negligible, although the QP gave more accurate results.

1) Studies the case in which the optimization strategy is based on splitting the problem (EV charging schedule) into smaller subproblems (multi-time slots).

2) No standard model, difficulty increases for complex problem.

3) Give global optimal result.



3) Maximize profit of provid-ing regulation services.

Stochastic programming Used in: [30], [42], [43], [44], [45].

The computation time is longer generally because more scenarios are considered.

The simulation results in [30] indicated that the difference between the power losses of the stochastic and the deterministic optimum is rather small.

Studies the case in which some of the constraints or parameters (Load profile, driving pattern etc.) depend on random varia-bles.



3) Maximize profit of provid-ing regulation services.

7 MATHEMATICAL MODELING AND CONTROL: DECENTRALIZED CONTROL

Compared to centralized control, the decentralized control is a relative new appli-cation to EV fleet control, but still a lot of efforts have been done considering the amount of the articles.

7.1 Two way price signal- Price and power negotiation As discussed in section 3, the following papers [7], [47]-[51] are chosen to further illustrate the two way price signal control method. Considering the similarities of the papers, we will discuss and present the papers in the following: the detailed formulas are given in the first paper with the purpose of facilitating reader’s under-standing. In paper [7], decentralized charging control of large population of electric vehicles is formulated as a class of finite-horizon dynamic games. Within this game, the control objective is to minimize electricity generation costs by establish-ing an EV charging schedule that fills the overnight demand valley. Moreover, the paper establishes a sufficient condition under which the system converges to the unique Nash equilibrium.

The key formulas are listed below:

𝑥𝑡+1𝑛 = 𝑥𝑡𝑛 +𝛼𝑛

𝛽𝑛𝑢𝑡𝑛, 𝑡 = 0, … ,𝑇 − 1

Where 𝑥𝑛 is the state of charge of 𝐸𝑉𝑛, 𝛼𝑛and 𝛽𝑛 means the charging efficiency and battery size of 𝐸𝑉𝑛, and 𝑢𝑛 represents the local control variable. The purpose of the study is to find the set of feasible full charging controls, which are described below:

𝜔𝑛 ≔ {𝑢𝑛 ≡ (𝑢0𝑛, … , 𝑢𝑇−1𝑛 ); 𝑠. 𝑡.𝑢𝑡𝑛 ≥ 0, 𝑥𝑇𝑛 = 1} Where the final constraint on 𝑥𝑇𝑛 requires that all EVs are fully charged by the end of the interval. The cost function of agent n, denoted by 𝐽𝑛(𝑢) is used as criteria and specified as:

𝐽𝑛(𝑢) ≔ �{𝑝(𝑟𝑡)𝑢𝑡𝑛 + 𝛿�𝑢𝑡𝑛 − 𝑎𝑣𝑔(𝑢𝑡)�2}

𝑇−1

𝑡=0

Where each agent’s optimal charging strategy must achieve a trade-off between the total electricity cost 𝑝(𝑟)𝑢𝑛 and the cost incurred in deviating from the average behaviour of the EV population�𝑢𝑛 − 𝑎𝑣𝑔(𝑢)�2. With these criteria and certain conditions, the theorem about the existence of the Nash equilibrium is presented in the paper.

The proposed algorithm ensures convergence to a flat, or optimally valley filling aggregate charging profile. However, in both papers [47], [48], all EVs are required to participated the negotiation at the same time, and implement the schedules they

commit to. In a more realistic scenario, EVs may join the negotiation at different time, not necessarily known to the FO beforehand. Furthermore, the approach is suitable to today’s system or those mainly comprised of conventional demand, this limits the cycling required in thermal plants; however the response to intermittent generation will be of more interest. During periods of intermittent (renewable) generation (RG), the price is unlikely to be directly related to demand, as RG typi-cally has lower or zero marginal cost.

Zhong Fan [50] applied the concept of congestion pricing in internet traffic control and showed that price information is very useful to regulate user demand and con-sequently balance the network load. Individual users adapt to the price signals to maximize their own benefits. User preference is modelled as a willingness to pay parameter which will influence both individual charging rate/cost and overall sys-tem behaviour, because, the unit price of energy in a time slot is a function of the aggregate demand in the paper. Charging power is allocated according to fair pay principle which is economically efficient and the mechanism ensure the system stable under arbitrary network topologies. However, the approach is not compatible with current market structure since in Zhong Fan’s paper, iterative convergence is required. Besides, the assumption that price is a function of demand, with a fixed constant of proportionality is weak, because reductions in demand won’t necessari-ly lead to corresponding reductions in price. Moreover, only the EV load is consid-ered in the paper, it is arguable to inquiry the conventional load. As mentioned by the authors in the paper, the proposed model is also a kind of game theory.

In short conclusion, both papers [49][50] made a good effort in trying to use game theory to formulate the complex decision making process for future energy traders, especially the FO. In the future smart grids, the distributed generation resources (DER) are most likely to be integrated via market-based mechanisms; therefore game theory will be a very useful tool to study the dual impacts between DERs and the markets.

7.2 One-way price signal-Price and demand elasticity By using one-way price signal, we mean that the EVs controller do not need to propose and submit their charging profile to the EV FO, instead the FOs will antic-ipate their response to the dynamic price. The dynamic price ranges from simple time-of-use electricity rate [52][53] to more varying hourly prices [54][55]. Both studies [52][53] suggested that the TOU rates can be properly designed to reduce the peak demand as EVs penetrate the vehicle market. However, it is also noted in [52] that the extent to which properly designed rates could assist in maintaining grid reliability will remain open until empirically tested EV owner’s price respon-siveness through experiment pilots are known.

Both studies [54][55] investigated the price elasticity of electricity consumers and these are also the key issues in one-way price signal approach. Details in [55] is

presented. In the model [55], the marginal utility function of loads is realized by the following parametric stochastic process:

( ), ;( )

0, .t t

r totherwise

β δ α α α γ− − ≤ ≤ +=

(4)

where , , ,α β γ δ are random variables that describes the different characteristics of utility function as follows:

a) α stands for the time slot that a task is initially requested, which also reflects the task distribution;

b) β is the initial marginal utility, which stands for the magnitude of the marginal utility.

c) γ is the tolerable delay, which determines the maximum delay that a user can tolerate to finish a task;

d) δ means the utility decay rate, which represents the cost of inconvenience by the delay.

Under this model, the scheduling of each individual task is now a random event whose probability distribution is controlled by the stochastic process 𝑟(𝑡). The aggregated demand curve can be estimated through expectation with respect to the distribution of 𝑟(𝑡). Note that some assumptions have been made before, such as the time period of the scheduling is divided into T time slots, the total M individual tasks m: m=1,…, M of different appliances that are to be initialized by all the users within the scheduling period, and each task will consumer 𝑥𝑚 kWh energy. Fur-thermore, it is assumed that each task can be completed within one time slot; there-fore, tasks that have duration longer than one time slot will be decomposed into multiple tasks that are considered independently.

In general, one can see that within decentralized control, no significant computing resources are required and the communication infrastructure is also simplified compare to centralized control.

8 CONCLUSION AND RECOMMENDATIONS

8.1 Conclusion and discussions As a conclusion, it is learned from this study that:

• Control objectives of aggregating a large penetration of EVs are essential for the starting of generating EVs’ schedules.

• Linear approximation of state of charge of battery (EV) is acceptable when doing the smart charging study.

• Linear programming is suitable for the smart charging study of EV fleet and individual EV.

• Price signal can be well designed and utilized to coordinate the charging profiles of EVs.

The following benefits of present study can be identified:

• The study outlines a foundation for future improvements in term of smart charging from a control theory perspective.

• The advantage and disadvantage of centralized and decentralized control are discussed, which gives a basis for comparing available methods for fu-ture developments.

• Details modelling method and algorithms are illustrated by showing the key formulas and compared in term of their performance, calculation time etc.

However, it should be observed and emphasized that the above discussion did not consider the real time operations, i.e., there is no continuous monitoring and as-sessment of the state of dynamic system and therefore lack of the appropriate re-sponse in abnormal situations. This means that new procedures considering the dynamic behaviour of EV fleet and distribution networks should be developed as well.

8.2 Recommendations on future research directions in the area Based on the discussions in the present study, future research directions are out-lined below:

1) Coordinate the multi-goals of smart charging of EVs

Recently, the trend in smart charging of EVs is to integrate the interests of EV owners, ancillary services required by the transmission system operator as well as respecting the hard constraint imposed by the distribution system operator. Re-search in [9] [21] aim to coordinate these multiple objectives centrally. Alternative-ly, some studies [56] used a price signal/market approach to coordinate the multi-ple objectives.

2) Integrating the control method

Although most research assumed either centralized control or decentralized control methods when starting the study, this is indeed an important decision which should be taken in the earlier stage. From our perspective, three issues shall be investigat-ed thoroughly.

• Depending on the aggregation goals, this is due to different goals have var-ious requirement on EVs in term of response time etc.

• Depending on the EV consumer’s participation, such as some consumers do not like their EVs to be controlled by FOs, under such circumstance, price incentives are a suitable method.

• Depending on the business model, we means whether the economic bene-fits of optimal charging of EVs can justify the cost of communication in-

frastructure in all cases; this will be an important consideration when choosing the control method.

Studies in [57], [58] compared the centralized control and decentralized control method when utilizing them to make an optimal plan which can optimal delivery energy to EVs as well as avoiding grid congestions. They outlined the advantages and disadvantages of both strategies. Fig. 5 illustrates the structures of integrating control strategies in a smart grid environment especially considering the congestion management in distribution networks. The argument for proposing this integration relies on the fact that this system architecture is comprehensive for the solution of integrating EVs into the power distribution systems.

Current market place

FO(Centralized)

FO(Decentralized)

EV EV EV

EVEVEV

Bids and activations

Grid Zone 1

Grid Zone 2

Direct control signal

Price signal

Figure 5: Integrating control method considering grid congestion man-

agement

3) A multi-agent systems based realization of smart charging of EVs

It is observed that when implementing both control strategies of smart charging of EVs, especially decentralized control method, multi-agents system based technolo-gy is very suitable to design a coordinated and collaborative system for an intelli-gent charging network of EVs. In the multi-agent systems, different interests of various actors shown in Fig.1 can be presented and coordinated by using smart charging method. By using multi-agent systems technology, one can model the optimizations and the negotiations happened in the smart charging of EVs. In [58],

[59], the authors modelled the smart charging of EVs using multi-agent systems technology.

9 ACKNOWLEDGEMENT The authors appreciate the valuable comments from the editor and the reviewers for improving the quality of the paper. Besides, we also would like to thank our colleague Dr. Peter Bach Andersen for his discussion on the communications be-tween the EVSEs and the EVs.

10 REFERENCES

[1] K. Parks, P. Denholm, and A. J. Markel, “Costs and emissions associated with plug-in hybrid electric vehicle charging in the Xcel Energy Colorado service territory”. National Renewable Energy Laboratory Golden, CO, 2007.

[2] C. Søndergren, Edison project: “Electric Vehicles in Future Market Models”, Danish Energy Association, 2011.

[3] R. J. Bessa and M. A. Matos, “Economic and technical management of an aggregation agent for electric vehicles: a literature survey,” European Transactions on Electrical Power, vol. 22, no. 3, pp. 334–350, 2012.

[4] T. G. San Román, I. Momber, M. R. Abbad, and Á. Sánchez Miralles, “Regulatory framework and business models for charging plug-in electric vehicles: Infrastructure, agents, and commercial relationships,” Energy Policy, vol. 39, no. 10, 2011.

[5] D. S. Callaway and I. A. Hiskens, “Achieving Controllability of Electric Loads” Proceedings of the IEEE, vol. 99, no. 1, 2011.

[6] K. Heussen, S. You, B. Biegel, L. H. Hansen, and K. B. Andersen,“Indirect control for demand side management - A conceptual introduction,” IEEE PES Innovative Smart Grid Technologies Conference Europe, pp. 1–8, 2012.

[7] Z. Ma, D. S. Callaway, and I. A. Hiskens, “Decentralized Charging Control of Large Populations of Plug-in Electric Vehicles,” IEEE Transactions on Control Systems Technology, vol. 21, no. 1, pp. 67–78, 2013.

[8] T. P. Lyon, M. Michelin, and A. Jongejan, “Is ‘smart charging’ policy for electric vehicles worthwhile?,” Energy Policy, vol. 41, 2012.

[9] J. A. P. Lopes, F. J. Soares, and P. M. R. Almeida, “Integration of electric vehicles in the electric power system,” Proceedings of the IEEE, vol. 99, no. 1, pp. 168–183, 2011.

[10] W. Kempton and J. Tomic, “Vehicle-to-grid power fundamentals: Calculating capacity and net revenue,” Journal of Power Sources, vol. 144, no. 1, pp. 268–279, 2005.

[11] W. Kempton and J. Tomic, “Vehicle-to-grid power implementation: From stabilizing the grid to supporting large-scale renewable energy,” Journal of Power Sources, vol. 144, no. 1, pp. 280–294, 2005.

[12] N. Rotering and M. Ilic, “Optimal Charge Control of Plug-In Hybrid Electric Vehicles In Deregulated Electricity Markets,” IEEE Transactions on Power Systems, 2010.

[13] S. Han and K. Sezaki, “Development of an optimal vehicle-to-grid aggregator for frequency regulation,” IEEE Transactions on Smart Grid, vol. 1, no. 1, pp. 65–72, 2010.

[14] C. K. Divya, J. Østergaard, E. Larsen, C. Kern, T. Wittmann, and M. Weinhold, “Integration of electric drive vehicles in the Danish electricity network with high wind power penetration,” European Transactions on Electrical Power, vol. 20, no. 7, pp. 872–883, 2010.

[15] F. Tuffner and M. Kintner-Meyer, “Using Electric Vehicles to Meet Balancing Requirements Associated with Wind Power,” Pacific Northwest National Laboratory (PNNL) , 2011.

[16] J. A. Pecas Lopes, P. M. Rocha Almeida, and F. J. Soares, “Using vehicle-to-grid to maximize the integration of intermittent renewable energy resources in islanded electric grids,” in Clean Electrical Power, 2009 International Conference on, 2009, pp. 290–295.

[17] J. A. P. Lopes, F. J. Soares, P. M. Almeida, and M. M. da Silva, “Smart Charging Strategies for Electric Vehicles: Enhancing Grid Performance and Maximizing the Use of Variable Renewable Energy Resources,” in EVS24-The 24th International Battery, Hybrid and Fuel Cell Electric Vehicle Symposium & Exhibition, Stavanger, Norway, 2009.

[18] H. Lund and W. Kempton, “Integration of renewable energy into the transport and electricity sectors through V2G,” Energy Policy, vol. 36, no. 9, pp. 3578–3587, 2008.

[19] J. Hu, S. You, J. Østergaard, M. Lind, and Q. W. Wu., “Optimal Charging Schedule of an Electric Vehicle Fleet,” in 46th Internation Univerisities Power Engineering Conference (UPEC), Germany., 2011.

[20] O. Sundström and C. Binding, “Optimization methods to plan the charging of electric vehicle fleets,” in Proceedings of the International Conference on Control, Communication and Power Engineering, 2010, pp. 28–29.

[21] O. Sundstrom and C. Binding, “Flexible Charging Optimization for Electric Vehicles Considering Distribution Grid Constraints,” IEEE Transactions on Smart Grid, vol. 3, no. 1, pp. 26–37, 2012.

[22] T. K. Kristoffersen, K. Capion, and P. Meibom, “Optimal charging of electric drive vehicles in a market environment,” Applied Energy, 2010.

[23] T. Lan, J. Hu, Q. Kang, C. Si, L. Wang, and Q. Wu, “Optimal control of an electric vehicle’s charging schedule under electricity markets,” Neural Computing and Application, 2012.

[24] Y. Cao, S. Tang, C. Li, P. Zhang, Y. Tan, Z. Zhang, and J. Li, “An Optimized EV Charging Model Considering TOU Price and SOC Curve,” IEEE Transactions on Smart Grid, vol. 3, no. 1, pp. 388–393, 2012.

[25] G. T. Heydt, “The Impact of Electric Vehicle Deployment on Load Management Strategies,” IEEE Power Engineering Review, vol. PER-3, no. 5, pp. 41–42, 1983.

[26] R. C. Green II, L. Wang, and M. Alam, “The impact of plug-in hybrid electric vehicles on distribution networks: A review and outlook,” Renewable and Sustainable Energy Reviews, vol. 15, no. 1, pp. 544–553, Jan. 2011.

[27] O. Sundstrom and C. Binding, “Planning electric-drive vehicle charging under constrained grid conditions,” in 2010 International Conference on Power System Technology (POWERCON), pp. 1–6.

[28] S. Letendre and R. A. Watts, “Effects of plug-in hybrid electric vehicles on the Vermont electric transmission system,” in Transportation Research Board Annual Meeting, Washington DC, 2009, pp. 11–15.

[29] J. A. P. Lopes, F. J. Soares, and P. M. R. Almeida, “Identifying management procedures to deal with connection of electric vehicles in the grid,” in 2009 IEEE Bucharest PowerTech, 2009, pp. 1–8.

[30] K. Clement-Nyns, E. Haesen, and J. Driesen, “The impact of charging plug-in hybrid electric vehicles on a residential distribution grid,” IEEE Transactions on Power Systems, vol. 25, no. 1, pp. 371–380, 2010.

[31] A. GOFUKU, “Representation of goals functions structures and derivation of behaviour for efficient design of engineering systems,” Technical University of Denmark, 1994.

[32] D. Peterson and J. Matute, “Electric vehicle service provider networks and market distortions,” 2012 26th Electric Vehicle Symposium, vol. 1, pp. 457–465, 2012.

[33] W. Su, H. Eichi, W. Zeng, and M.-Y. Chow, “A Survey on the Electrification of Transportation in a Smart Grid Environment,” IEEE Transactions on Industrial Informatics, vol. 8, no. 1, pp. 1–10, 2012.

[34] S. Käbisch, A. Schmitt, M. Winter, and J. Heuer, “Interconnections and Communications of Electric Vehicles and Smart Grids,” in 2010 First IEEE International Conference on Smart Grid Communications (SmartGridComm), IEEE, 2010., 2010, pp. 161–166.

[35] J. Schmutzler and C. Wietfeld, “Analysis of message sequences and encoding efficiency for electric vehicle to grid interconnections,” in 2010 IEEE Vehicular Networking Conference, 2010.

[36] J. Kirby and F. Hassan, “AC Recharging Infrastructure for EVs and future smart grids — A review,” in 2012 47th IEEE International Universities Power Engineering Conference (UPEC), 2012., 2012, pp. 1–6.

[37] J. Fluhr and T. Lutz, “Use case types for communication with and for Electric Vehicles (EV),” in 17th Internation Conference on Concurrent Enterprising (IEC 2011), 2011, pp. 1–6.

[38] H. Bindner, T. Cronin, P. Lundsager, J. F. Manwell, U. Abdulwahid, and I. Baring-Gould, “Lifetime modelling of lead acid batteries,” Benchmarking, vol. 12, p. 82, 2005.

[39] L. Guzzella and A. Sciarretta, Vehicle propulsion systems: introduction to modeling and optimization. Springer Verlag, 2007.

[40] J. Nickoletatos and S. Tselepis, “Results and analysis of simulated cycling tests on batteries,” CRES report (for Benchmarking project), 2004.

[41] S. Shahidinejad, E. Bibeau, and S. Filizadeh, “Statistical Development of a Duty Cycle for Plug-in Vehicles in a North American Urban Setting Using Fleet Information,” IEEE Transactions on Vehicular Technology, vol. 59, no. 8, pp. 3710–3719, 2010.

[42] J. Fluhr, K. H. Ahlert, and C. Weinhardt, “A stochastic model for simulating the availability of electric vehicles for services to the power grid” in 43th International Conference on Systems Sciences, Hawaii, 2010, pp. 1–10.

[43] W. Su and M. Y. Chow, “Evaluation on intelligent energy management system for PHEVs/PEVs using Monte Carlo method,” in 2011, 4th International Conference on Electric Utility Deregulation and Restructuring and Power Technologies (DRPT), 2011, pp. 1675–1680.

[44] J. Taylor, A. Maitra, M. Alexander, D. Brooks, and M. Duvall, “Evaluation of the impact of plug-in electric vehicle loading on distribution system operations,” in Power & Energy Society General Meeting, 2009. PES’09. IEEE, 2009, pp. 1–6.

[45] P. Papadopoulos, I. Grau, L. M. Cipcigan, and N. Jenkins, “Predicting Electric Vehicle Impacts on Residential Distribution Networks with Distributed Generation,” in 2010 IEEE Vehicle Power and Propulsion Conference, 2012.

[46] J. Linderoth, A. Shapiro, and S. Wright, “The empirical behavior of sampling methods for stochastic programming,” Annals of Operations Research, vol. 142, no. 1, pp. 215–241, 2006.

[47] Z. Ma, D. Callaway, and I. Hiskens, “Decentralized charging control for large populations of plug-in electric vehicles,” in Decision and Control (CDC), 2010 49th IEEE Conference on, 2010, pp. 206–212.

[48] L. Gan, U. Topcu, and S. Low, “Optimal Decentralized Protocol for Electric Vehicle Charging,” IEEE Transactions on Power Systems, vol. 28, no. 2, pp. 940–951, 2013.

[49] S. Walid, H. Zhu, V. P. H, and B. Tamer, “A Noncooperative Game for Double Auction-Based Energy Trading between PHEVs and Distribution Grids,” in Proceedings of IEEE SmartGridComm2011, 2011.

[50] Z. Fan, “Distributed Charging of PHEVs in a smart Grid,” in Proceedings of IEEE SmartGridComm2011, 2011.

[51] Q. Li., T. Cui, N. Rohit, F. Franz, and M. Ilic., “On-Line Decentralized Charging of Plug-In Electric Vehicles in Power System,” arXiv, 2011.

[52] S. Shao, T. Zhang, M. Pipattanasomporn, and S. Rahman, “Impact of TOU rates on distribution load shapes in a smart grid with PHEV penetration,” 2010 IEEE PES Transmission and Distribution Conference and Exposition: Smart Solutions for a Changing World, 2010.

[53] A. Faruqui, R. Hledik, A. Levy, and A. Madian, “Will Smart Prices Induce Smart Charging of Electric Vehicles?,” The Brattle Group Discussion, 2011.

[54] O. Corradi, H. Ochsenfeld, H. Madsen, and P. Pinson, “Controlling Electricity Consumption by Forecasting its Response to Varying Prices,” IEEE Transactions on Power Systems, vol. 28, no. 1, pp. 421–429, 2013.

[55] R. Yu, W. Yang, and S. Rahardja, “A Statistical Demand-Price Model With Its Application in Optimal Real-Time Price,” IEEE Transactions on Smart Grid, vol. 3, no. 4, pp. 1734–1742, 2012.

[56] P. Bach Andersen, J. Hu, and K. Heussen, “Coordination strategies for distribution grid congestion management in a multi-actor, multi-objective setting,” IEEE PES Innovative Smart Grid Technologies Conference Europe, pp. 1–8, 2012.

[57] P. Richardson, D. Flynn, and A. Keane, “Local Versus Centralized Charging Strategies for Electric Vehicles in Low Voltage Distribution Systems,” IEEE Transactions on Smart Grid, vol. 3, no. 2, pp. 1020–1028, 2012.

[58] E. L. Karfopoulos and N. D. Hatziargyriou, “A Multi-Agent System for Controlled Charging of a Large Population of Electric Vehicles,” IEEE Transactions on Power Systems, vol. 28, no. 2, pp. 1196–1204, 2013.

[59] J. Miranda, J. Borges, D. Valério, and M. J. G. C. Mendes, “Development of a multi-agent management system for an intelligent charging network of electric vehicles,” IFAC Proceedings Volumes, vol. 18, no. PART 1, pp. 12267–12272, 2011.

Optimal charging schedule of an electric vehicle fleet 103

A.2 Optimal charging schedule of an electric ve-hicle fleet

This paper was presented at 46th International Universities’ Power EngineeringConference (UPEC), Spet 2011, Soest, Germany.

Optimal Charging Schedule of an Electric VehicleFleet

Junjie Hu, Shi You, Jacob Østergaard, Morten Lind and QiuWei WuCenter for Electric Technology

Technical University of DenmarkKgs.2800, Lyngby, Denmark

Email: {junhu, sy, joe, mli, qw }@elektro.dtu.dk

Abstract—In this paper, we propose an approach to optimizethe charging schedule of an Electric Vehicle (EV) fleet both takinginto account spot price and individual EV driving requirementwith the goal of minimizing charging costs. A flexible and suit-able mathematic model is introduced to characterize the smartcharging behavior and detailed parameters needed for chargingbehavior of an individual EV are analyzed. The individualcharging schedule is extended to the EV fleet. Simulation resultsare presented to illustrate the effectiveness of the proposed model.

Index Terms—Electric Vehicle, Fleet Operator, ChargingSchedule, Linear Program

I. INTRODUCTION

The Danish government has issued a long-term energypolicy aiming at 30% production from renewable energy and50% of the average electricity consumption from wind powerby 2025 [1]. This policy conforms well to the procedure ofdealing with global environmental issues and has led to agrowing interest in EVs. EVs have V2G and G2V capabil-ities which can be used as a storage device for smoothingwind power fluctuations and to provide more reliable poweroperation as well as presents flexible demands.

Several investigations have been made to reach the goalof this new Danish energy policy, Larsen et al. [2] proposedscheme to improve system operations, which considered EVsas new power consumers and also the possibilities as providersof energy storage. The possibility of using vehicle to gridto improve wind power integration was studied in [3] andthe result suggested that a small share of EVs may partiallydiminish the short period oscillations, while a high shareof EVs achieved much better improvements. A feasibilitystudy of implementing V2G scenario in Denmark was done in[4][5]. The system constraints for integrating EVs into powersystems were examined in the study and the technical andeconomical viability of various possible V2G architectureswere evaluated. It was concluded that the V2G technology canassist in realizing the goal of Danish government. In additionalto these feasibility studies, a V2G demonstration project wasimplemented by AC Propulsion Inc. to evaluate the practicalityof EVs providing regulation services [6].

Extensive research about EV optimal charging managementhave been conducted around the world in recent years. As earlyas 1983, Heydt [7] researched the impact of EV deployment

on loads and proposed the methods to alleviate peak loadingwith off-peak recharging and the shift of peak loads to off-peak periods. Koof et al. [8] presented an extensive study oncontrolling an EV energy system to reduce the fuel consump-tion and exhaust emissions with three algorithms, dynamicprogramming, quadratic programming and model predictivecontrol. Kristien et al. [9] researched the impact of chargingPHEVs on a residential distribution grid and illustrated theresults of coordinated and uncoordinated charging. Niklas etal. [10] proposed two algorithms to address optimal chargingcontrol to avoid the peak load, one is to optimize the chargingtime and energy flow, another is generating profits by partici-pating ancillary service markets which can alleviate the peakload. Sundstrom and Binding [11] established charging plan tooptimize the cost as well as achieve optimal power balancing.In [12], Kristoffersen presented a framework for optimizingelectric drive vehicles charging and discharging given vari-ations in electricity spot prices and driving patterns of thevehicle fleet. However, some methods lack the flexible abilityto adapt to any driving pattern since most of these methodsare based on an hourly charging scenario, some methods didnot present a simple and effective way to formulate and solvethe optimal charging problem of EV fleets, and most of theminvestigated the problem from the overall availability of theEV fleet instead of from the individual EVs perspective.

The purpose of this paper is to investigate methods ofobtaining the optimally individual EVs charging schedule andthe charging schedule of an EV fleet and test it with realisticdata. Detailed model is used to address the individual chargingschedule and is then extended to an EV fleet. Based onpredicted data including driving pattern information [13] andelectricity spot price, the day-ahead charging schedule of anEV fleet is obtained by linear programming. The paper isorganized as follows: Section II describes the system archi-tecture of an fleet operator and some assumptions. In sectionIII, we present methods to formulate the problem and get theindividual charging schedule. The methods are extended to alarge scale of an EV fleet in section IV. Simulation resultsare illustrated in Section V. Finally, concluding remarks arecollected in Section VI.

Fig. 1. System architecture of fleet operator

II. SYSTEM ARCHITECTURE AND ASSUMPTIONS

Nowadays, it is widely accepted to charge the EV batterieswith mainly three options [14]:

1. Fast charging: pull over EVs at the fast charging stationsand charge the EV batteries within, e.g., 10-15 minutes.

2. Battery swapping: drive EVs to the battery swappingstations and exchange the used battery with a fully chargedone.

3. Low power charging: charge EVs at homes, parking lotsnear home, working place or shopping malls.

The third option is analyzed for this paper, by introducingmore advanced communication equipments, a centralized con-trol system fleet operator is used for the interactions betweenthe EVs and the electricity market. The overview systemarchitecture of the fleet operator is depicted in Fig.1. The fleetoperator need historical or statistic data of all the EVs andelectricity spot price in order to predict spot price and drivingpattern for next day. In this paper, we will not investigate theway of forecasting these data, the fleet operator is assumed toobtain the predicted electricity spot price and driving pattern.In term of relevance of the predicted electricity spot price, theNordic electricity market is ideally suited for the applicationof optimal charging control with fleet operator for its day-ahead spot market price. Some researches [15][16][17] havebeen done on forecasting the electricity price in the Nordicelectricity market. Thus it is reasonable for the fleet operatorto predict the spot price for next day and to manage the optimalcharging behavior according to user’s preference.

Some assumptions are given for the fleet operator beforewe start to formulate and analyze the problem:

• Based on a previous study [11], the difference is relativelysmall between of charging schedule based on a linear approx-imation battery model and a quadratic approximation batterymodel. A Linear battery model is chosen for simplicity in thispaper.

• Charging power is constant.• The economic impacts of depth of discharge (DOD) to

the battery life are ignored.• Energy content E0, i.e. initial status of battery SOC0 is

assumed to be known, the state of charge (SOC) is the relativeenergy level

SOC =Et

Ecap.100%

where Et is the available energy of the battery in time t, Ecap

is the rated capacity of the battery.

III. METHODS

In this section, we analyze the optimal charging processwith the goal of minimizing the charging cost when getting theday-ahead charging schedule of an EV fleet. The individual’soptimal charging process is considered first since the chargingschedule should be in accordance to their requirements. Withthe constant charging power ability, this problem can beformulated as a linear one:

minCT E

subject to

AtE ≥ Be

AtE ≤ Bb

AtE ≥ Bset

0 ≤ E ≤ Emax

(1)

with the cost vector C, the decision variable vector E, the stop-over inequality constraints (At, Be), the battery inequalityconstraints (At, Bb), the inequality constraints (At, Bset) andthe upper and lower bounds (0, Emax). The first constraintmeans that the available energy in battery should be greaterthan or equal to the energy requirement for the next trip. Thesecond constraint indicates that available energy in batteryshould be less than or equal to the maximum power capacity.The third constraint requires that the planning charge energyshould accord to the user’s special preference. The last onerepresents that the planning charge energy should less than itscharging ability.

Assume j = 1, 2, ..., n, ..., N is the index for the time slotcontained in one plan duration. k = 1, 2, ..., mj is the indexfor the stop-over contained in one time slot j. Therefore, thecost vector C comprises the cost concerning the time slot jand stop-over index k, Cj,k. The charging energy vector Ecomprises the charging energy for each time slot index andstop-over index Ej,k.

C =

[C1,1, C1,2, ..., C1,m1 , ..., CN,1, CN,2, ..., CN,mN

]T

(2)

E =

[E1,1, E1,2, ..., E1,m1 , ..., EN,1, EN,2, ..., EN,mN

]T

(3)As we assumed before, individual’s driving pattern is pre-

dicted, which means the following information is available foreach EV:

⋄ Time of stop-over tstopj,k

⋄ Time of disconnection (driving time) tdrivej,k, the k isthe same one with the stop-over index.

⋄ The time loss during the process of connecting the car tothe grid is ignored when the driving procedure is finished andthe connection time is assumed equal to the stop-over time.

⋄ Energy requirement for the next trip Edrivej,k. This canbe calculated with the driving time, driving speed and energyconsumption per km. An example is given in section VI.

The minimum amount of energy that need to be charged inone time slot j before the end of stop-over k should requirethe following inequality

SOC0 +

n∑

j=1

r∑

k=1

Ej,k ≥ SOCMin +

n∑

j=1

r∑

k=1

Edrivej,k, (4)

where SOCMin means the lowest allowed SOC of EV battery,denotes Min, r is the rth stop-over during each time slot j.

The stop-over inequality constraints At is a lower triangularunit matrix and the dimensions of matrix At can be deducedfrom the time slot j and stop-over index k.

From inequality formula 4, it follows that

Be =

[{−E0 + Min + Edrive1,1}, ...,

{−E0 + Min + Edrive1,1 + ... + Edrive1,m1}, ...,

{−E0 + Min + Edrive1,1 + ... + Edrive1,m1 + ... +

EdriveN,1}, ..., {−E0 + Min + Edrive1,1 + ... +

EdriveN,1 + ... + EdriveN,mN}]T

(5)

Let the maximum amount charged energy in each stop-overk during slot j be

SOC0 +n∑

j=1

r∑

k=1

Ej,k ≤{

wEcap +n∑

j=1

r∑

k=1

Edrivej,k−1

},

(6)

where w is the parameter which express that the chargingbehavior of the battery is a linear process.

Thus, one can get

Bb =

[{wEcap − E0}, ...,

{wEcap − E0 + Edrive1,1... + Edrive1,m1−1}, ...,

{wEcap − E0 + Edrive1,1... + Edrive1,m1−1... +

EdriveN,1}, ..., {wEcap − E0 + Edrive1,1... +

Edrive1,m1 ... + EdriveN,1... + EdriveN,mN −1}]T

(7)

In reality, it is reasonable to assume that users expect theEV battery at a certain status in a specific time. This can beillustrated by the following equation:

0

1

2

3

4

5

6

7

8

0:00 1:00

2:00 3:00

4:00 5:00

6:00 7:00

7:10 7:50

8:00 9:00

10:00 11:00

12:00 13:00

14:00 15:00

16:00 16:10

16:50 17:00

18:00 19:00

20:00 21:00

22:00 23:00

Energy/kWh

Time/h

Extra!driving!requirements!

in!Scenario!B

Basic!driving!requirements!

in!scenario!A

Fig. 2. Driving requirements Edrivej,k for EV 1

n∑

j=1

r∑

k=1

Ej,k + SOC0 −n∑

j=1

r∑

k=1

Edrivej,k ≥ SOCset (8)

Because the charging power ability is constant, we letEmax = P ∗ tstopj,k.

IV. EXTENSION TO A LARGE SCALE OF EV FLEET

Since the model is suitable for each vehicle, and if matrixof At is noted by Aj,k in formula (1), i is introduced to denotethe index of EV, 1, 2, ..., i, ..., M , then the extendable modelof EV fleet can be given by following formula:

A1,j,k

A2,j,k

...AM,j,k

(9)

The same approach is also used for other parameters Be, Bb. Inthis way, the charging schedule for each EV can be obtained.For individual EVs, the charging schedule may include twoor more charged energy sections in one time slot. With thepurpose of calculating the whole charged energy/power in onetime slot, we will sum those time slots in which they containtwo or more charged energy section for each individual EV:

Ei,j =

mj∑

k=1

Ei,j,k (10)

We can then get the charging schedule of the EV fleet by thesum of each EV corresponding to each time slot, which canbe expressed by the following equation:

Ej =

M∑

i=1

Ei,j . (11)

V. RESULTS

In this section, we give examples to demonstrate the ef-fectiveness of the proposed methods with two scenarios. Inscenario A, two cases are studied with the goal of minimizingthe charging cost. In order to manage the uncertainty ofthe driving pattern, a certain amount of driving requirementsare added for each Edrivej,k in scenario B. Besides this,

0

1

2

3

4

5

6

7

8

0:00 1:00

2:00 2:10

2:30 3:00

4:00 5:00

6:00 7:00

7:10 7:50

8:00 9:00

10:00 11:00

12:00 13:00

14:00 15:00

16:00 16:10

16:50 17:00

18:00 19:00

19:40 20:20

21:00 22:00

23:00 24:00

Energy/kWh

Time/h

Extra!driving!requirements!in!

Scenario!B

Basic!driving!requirements!in!

scenario!A

Fig. 3. Driving requirements Edrivej,k for EV 2

the optimal charging results concerning the user’s specialpreference are also studied. In addition, longer time horizonsare introduced for planning charging with the purpose ofgetting more optimal results. All three studies in scenario Bare in terms of other factors.

A. Optimal Charging Schedule With Basic Driving Require-ments

This scenario presents the case studies with basic drivingrequirements. The plan duration is 24 hours and the time slot isset to 1h. We expect to use the representative driving patternsof EVs in Denmark to test and illustrate this methods, however,this kind of driving pattern are not available at this moment.But, we are in the process of generating this driving patternfrom the GPS survey data on vehicle fleets (December 2001-April 2003) provided by Department of Transport, TechnicalUniversity of Denmark. Meanwhile, this paper focus on themethods to generate the charging schedule for an EV fleet,therefore, some artificial driving requirements and assumptionsare given in the following:

• Based on the study [13], 150Wh/km is used to calculatethe energy consumption from the driving distance.

• The drive speed is constant for all EVs, e.g. 60km/h.• Power capacity of Battery is 20kWh, i.e. Ecap and the

charging power is 2.3kW based on 10A ∗ 230V .• State of charge (SOC) should be between 20% and 85%

Ecap, because the battery life can be longer if the SOC is largethan or equal to 20% and the charging behavior will be closeto linear equation when the SOC is below 85%.

Two individual cases EV 1 and EV 2 are consideredbased on the above assumptions, representing the normal andcomplicated case, respectively. In Fig. 2 and Fig. 3, we givethe driving requirements for these two EVs with the light blueblock. Fig. 4 presents the spot price in one day (blue curve)(12, Jan, 2011), DK-West, from NordPool, which is used forthe predicted price for fleet operator.

The initial energy state E0 is set 4kWh, 4kWh for EV 1and EV 2, respectively. In Fig. 5 and Fig. 6, the chargingschedule for EV 1 and EV 2 are given. It is clear that thecharging time is located during the time slots where theelectricity price is lower compared to other time slots whilestill fulfilling the driving needs.

0.00

100.00

200.00

300.00

400.00

500.00

600.00

1 3 5 7 9 11 13 15 17 19 21 23

DKK/MWh

Time/h

First Day

SecondDay

Fig. 4. Two days Spot price

0.50

0.00

0.50

1.00

1.50

2.00

2.50

1 2 3 4 5 6 7 8 9 101112 13141516 17 18192021 222324

Energy/kWh

Time/h

Fig. 5. Charging schedule of EV1 in scenario A

0.00

0.50

1.00

1.50

2.00

2.50

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Energy/kWh

Time/h

Fig. 6. Charging schedule of EV2 in scenario A

1.00

0.00

1.00

2.00

3.00

4.00

5.00

1 3 5 7 9 11 13 15 17 19 21 23

Energy/kWh

Time/h

EV2

EV1

Fig. 7. Charging schedule of EV fleet in scenario A

0 5 10 15 20 250

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Time/h

SO

C%

Fig. 8. State of Charge of EV1 in scenario A

0.00

0.50

1.00

1.50

2.00

2.50

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47

Energy/kWh

Time/h

Fig. 9. Charging schedule of EV1 within longer plan duration in scenario B

B. Optimal Charging Schedule With other Factors

The question in this mathematical model is that this willalways result in minimum SOC at the end of the plan duration,which can be shown by the fig. 8. In order to manage it,the extended plan duration is introduced here. By takingadvantage of more decision variables, more optimal results canbe obtained. One question remains: how long should the planduration be? This relies on the prediction of the spot price. Ifthe more accurately predicted spot price is available for thislonger plan duration, the fleet operator can benefit from suchsolution. However, the fleet operator might lose more if thereare large deviations between forecasted and the real price. Anexample with a two-days time duration is studied here, Fig. 4shows continuous two-day electricity spot price, (12-13, Jan,2011), DK-West. The charging schedule is shown in Fig. 9,in which the last hour of the first day is also charged for nextday. With this method, the day-ahead charging schedule canbe iteratively calculated.

In addition, the above scenario has limitations when theusers might want to change their driving requirements, e.g,need more energy than predicted Edrivej,k. Therefore, with

0

0.5

1

1.5

2

2.5

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Energy/kWh

Time/h

Fig. 10. Charging schedule of EV1 with more driving requirements inscenario B

the purpose of managing the uncertainty of user’s drivingrequirements, it is reasonable to add a certain amount driv-ing requirements for the predicted Edrivej,k. But how manydriving requirements should be added? This question shouldrely on two main aspects, charging costs and the possibilitiesof managing uncertainties of driving requirements. In otherwords, the charging costs can be lower with the more accu-rately predicted driving requirements. We do not give such dataanalysis since the real driving pattern is not available now. Onecase is given to show the difference, this can be illustrated inFig 2 and Fig 3 with the red block on the top of the light blueblock, which here 20% more driving requirements are addedfor each Edrivej,k. The corresponding charging schedule isshown by Fig 10.

Besides the above conditions, some users might have specialpreferences, e.g, they want to ensure their EVs are close to befully charged in the morning since they can have more flexiblechoices. If we assume the users in Scenario A are risk takers,then the users in this Scenario are more focus on the reliability.However, how far the SOC of the battery should be charged?This is the similar problem compared to the above one. Also,we give an example, based on the normal case EV1 with thelight blue in Fig. 2 driving requirements. We assume that thisuser want his/her EV to be charged to SOC 85% at 7:00 AM.Fig. 11. illustrates the charging schedule.

The above case studies only gave examples of an EV fleetwith two EVs and Fig. 7 illustrates the result of EV fleetcharging schedule. This is indeed a small number. Fig. 12,which is obtained from ongoing project named Edison [18],can illustrate the aggregation management of the EV fleet byfleet operator. Meanwhile, this picture also show the potentialimportance of the fleet operator, since a large scale EV fleetneed to be managed which can optimally use the batteries tobalance the intermittency of wind. We think the above workcan give a method for the fleet operator to obtain the optimalcharging schedule for the whole EV fleet.

0.00

0.50

1.00

1.50

2.00

2.50

1 2 3 4 5 6 7 8 9 101112131415161718192021222324

Energy/kWh

Time/h

Fig. 11. Charging schedule of EV1 with user’s special preference in scenarioB

Fig. 12. Demo aggregating EV fleet

VI. DISCUSSION AND CONCLUSIONS

This paper presents a theoretical framework for designingand analysis of an EV fleet optimal charging schedule. Themodel reveals a general charging process for individual EVsto get their charging schedules with the criteria of obtaininga minimum charging cost. By formulating this problems vialinear models under a certain assumptions, this model foroptimal charging can adapt to any driving pattern with flexibletime intervals (from hour to minutes).

Future research is planned to be done from the followingperspectives:

1. Analyze the given ’Danish National Transport Surveydata’ and present a representative driving pattern, and thisdriving pattern will be used in the analysis of EV optimalcharging.

2. Extension the EV fleet charging schedule with intra-day markets, regulating power markets. With the intra-daymarkets and regulating power markets, the EV aggregator candynamically evaluate the day-ahead charging schedule andobtain the optimal potential profits after day-ahead market.This research can also be extended to a real time market.

ACKNOWLEDGMENT

The authors would like to thank the finical support of theDanish FORSKEL program. This project named ”EDISON” ,

currently undertaken at the Center for Electric Technology,Technical University of Denmark and partner contributorswithin the program, which aims to utilize the EV as balancingresources to support the long term goal of integrating 50%wind power capacity [5][18]. The authors also would like togive thanks to Anders Bro Pedersen for his comments.

REFERENCES

[1] Z. Xu, M. Gordon, M. Lind, and J. Ostergaard, “Towards a Danish powersystem with 50% windłSmart grids activities in denmark,” in Power &Energy Society General Meeting, 2009. PES’09. IEEE, pp. 1–8, IEEE,2009.

[2] E. Larsen, D. Chandrashekhara, and J. Ostergard, “Electric vehicles forimproved operation of power systems with high wind power penetra-tion,” in Energy 2030 Conference, 2008. ENERGY 2008. IEEE, pp. 1–6,IEEE, 2009.

[3] J. Brassin, “Vehicle to grid improving wind power integration,” Master’sthesis, Center for Electric Technology, Technical University of Denmark,2007.

[4] D. Chandrashekhara, J. Horstmann, J. Østergaard, E. Larsen, C. Kern,T. Wittmann, and M. Weinhold, “Vehicle to Grid (V2G) in Denmark-Feasibility Study,” tech. rep., Technical University of Denmark DK-Kgs.Lyngby, 2008.

[5] J. Østergaard, A. Foosnæs, Z. Xu, T. Mondorf, C. Andersen,S. Holthusen, T. Holm, M. Bendtsen, and K. Behnke, “Electric vehiclesin power systems with 50% wind power penetration: the Danish caseand the EDISON programme,” in European Conference Electricity &Mobility.

[6] A. Brooks, A. Propulsion, and C. A. R. B. R. Division, Vehicle-to-grid demonstration project: Grid regulation ancillary service with abattery electric vehicle. California Environmental Protection Agency,Air Resources Board, Research Division, 2002.

[7] G. Heydt, “The impact of electric vehicle deployment on load manage-ment straregies,” IEEE Transactions on Power Apparatus and Systems,no. 5, pp. 1253–1259, 2007.

[8] M. Koot, J. Kessels, B. De Jager, W. Heemels, P. Van den Bosch,and M. Steinbuch, “Energy management strategies for vehicular electricpower systems,” IEEE transactions on vehicular technology, vol. 54,no. 3, pp. 771–782, 2005.

[9] K. Clement-Nyns, E. Haesen, and J. Driesen, “The impact of chargingplug-in hybrid electric vehicles on a residential distribution grid,” PowerSystems, IEEE Transactions on, vol. 25, no. 1, pp. 371–380, 2010.

[10] N. Rotering and M. Ilic, “Optimal charge control of plug-in hybridelectric vehicles in deregulated electricity markets,” IEEE Transactionson Power Sy, 2010.

[11] O. Sundstrom and C. Binding, “Optimization methods to plan thecharging of electric vehicle fleets,” in Proceedings of the InternationalConference on Control, Communication and Power Engineering, pp. 28–29.

[12] T. Kristoffersen, K. Capion, and P. Meibom, “Optimal charging ofelectric drive vehicles in a market environment,” Applied Energy, 2010.

[13] Q. Wu, A. Nielsen, J. Østergaard, S. Cha, F. Marra, Y. Chen, andC. Træholt, “Driving Pattern Analysis for Electric Vehicle (EV) GridIntegration Study,” IEEE Inovative Smart Grid Technology Europe 2010.

[14] Q. Wu, A. Nielsen, J. Østergaard, S. Cha, F. Marra, and P. Andersen,“Modeling of Electric Vehicles (EVs) for EV Grid Integration Study,”in European Conference SmartGrids & E-Mobility 2nd.

[15] G. Koreneff, A. Seppala, M. Lehtonen, V. Kekkonen, E. Laitinen,J. Hakli, and E. Antila, “Electricity spot price forecasting as a partof energy management in de-regulated power market,” in Energy Man-agement and Power Delivery, 1998. Proceedings of EMPD’98. 1998International Conference on, vol. 1, pp. 223–228, IEEE, 2002.

[16] J. Lucia and E. Schwartz, “Electricity prices and power derivatives:Evidence from the nordic power exchange,” Review of DerivativesResearch, vol. 5, no. 1, pp. 5–50, 2002.

[17] I. Vehvilainen and T. Pyykkonen, “Stochastic factor model for electricityspot price–the case of the Nordic market,” Energy Economics, vol. 27,no. 2, pp. 351–367, 2005.

[18] E. Project(2009) in [Online].www.edison-net.dk.

110 Core publications

A.3 Optimal control of an electric vehicle’s charg-ing schedule under electricity markets

This paper is published at Neural Computing and Applications, Volume 23,Issue 7-8, pp 1865-1872, Dec 2013.

ISNN2012

Optimal control of an electric vehicle’s charging scheduleunder electricity markets

Tian Lan • Junjie Hu • Qi Kang • Chengyong Si •

Lei Wang • Qidi Wu

Received: 16 April 2012 / Accepted: 11 September 2012

� Springer-Verlag London Limited 2012

Abstract As increasing numbers of electric vehicles

(EVs) enter into the society, the charging behavior of EVs

has got lots of attention due to its economical difference

within the electricity market. The charging cost for EVs

generally differ from each other in choosing the charging

time interval (hourly), since the hourly electricity prices are

different in the market. In this paper, the problem is for-

mulated into an optimal control one and solved by dynamic

programming. Optimization aims to find the economically

optimal charging solution for each vehicle. In this paper, a

nonlinear battery model is characterized and presented, and

a given future electricity prices is assumed and utilized.

Simulation results indicate that daily charing cost is

reduced by smart charing.

Keywords Optimal control � Dynamic programming �Nonlinear control � Electric vehicles

1 Introduction

An important solution to curb CO2 emission and oil depen-

dency taken by the automobile industry is the introduction of

electric vehicles (EVs) [1, 2], since the EVs can shift

petroleum consumption to electricity. As an asset, it is well

understand that the EV can provide valuable service to power

systems, more than its transportation function. On the one

hand, battery of the EV can be considered as a controllable

load. With optimal charging or smart charging for the bat-

tery, the peak load can be shaved, and by doing this, EV

owners could maximize their profits by purchasing energy at

the possibly lowest electricity price. On the other hand,

battery of the EV can also be seen as energy storage equip-

ment which has possibility to provide vehicle-to-grid (V2G)

and grid-to-vehicle service and thereby earning profit [3–6].

Therefore, an optimal charging scheme is required to coor-

dinate the needs from the power system and maximize the

EV owner’s profit. In a general word, for a power system

operator, ‘‘optimal’’ can be interpreted as enable the large

penetration of renewable and distributed energy resources,

like wind power, photovatics, and also accommodate the

new loads, like EV and heat pump, both reliably; from EV

users perspective, optimal means minimizing their charging

cost without interfering their daily drive profile.

Many researches have been done on EV optimal charging

management. In the early 1980s, Heydt has already resear-

ched the impact of electric vehicles on the grid and con-

cluded that typical driving patterns will likely to coincide the

charging with peak load periods of power system [7]. So,

methods are developed to avoid overloading with off-peak

charging. In paper [8], Kristien researches the impact of

charging PHEVs on a residential distribution grid, investi-

gates the difference between coordinated and uncoordinated

charging with respect to various penetrations of PHEVs. Olle

presents a linear approximation-based method to formulate

and coordinate the optimal charging problems, grid con-

straints in terms of thermal overloading are considered. With

this method, the flexible charging ability of EVs is utilized to

T. Lan

Institute of Sustainable Electric Networks and Sources

of Energy, Technical University of Berlin, 10587 Berlin,

Germany

J. Hu

Center for Electric Technology, Technical University

of Denmark, 2800 Copenhagen, Denmark

Q. Kang (&) � C. Si � L. Wang � Q. Wu

Department of Control Science and Engineering,

Tongji university, Shanghai 201804, People’s Republic of China

e-mail: [email protected]

mitigate the grid overloading problem; and with this charg-

ing flexibility, individual EV owner’s can be controlled to

charge at the period with the lower electricity price [9]. In

paper [10], Niklas proposes two dynamic programming-

based algorithms to find the economically optimal solution

for vehicle owner. The first reduces daily electricity cost

substantially. The latter takes into account vehicle-to-grid

support as a means of generating additional profits by par-

ticipating in ancillary service markets. Sekyung proposes an

aggregator that makes efficient use of the distributed power

of electric vehicles to produce the desired grid scale power,

which is V2G concept that can make revenue from providing

regulation service [3].

Typically, an optimal control problem includes a cost

function of state and control variables and the control

objective is to find the paths of the control variables that

minimize the cost functions. In the scope of smart charging

of EVs, the cost function can be derived from two per-

spectives: one is the single charging cost of EV without

grid constraint, which means the grid is large to handle the

new load of EVs. Another optimal control problem is to

minimize the total monetary cost of fulfilling the charging

requirement, given assume spot price and monetary price

for overloading phenomenon. In this paper, we take the

first option and consider the EV as a controllable load, and

investigate its smart charging potential. The functionality

of regulation service will not be discussed here.

The following paper is organized as follows. A nonlin-

ear model of electric vehicle is modeled in Sect. 2. Section

3 gives a system architecture with appropriate assumptions.

Section 4 constructs a dynamic programming-based

mathematical model. In Sect. 5, one case is studied to

investigate the optimization of EV charging cost.

2 Electric vehicle model

With the purpose of charging planning, the EVs are con-

sidered to be battery packs in this study that have nonlinear

behavior [9]. Each battery is modeled as a steady-state

equivalent circuit, which represented by an ideal voltage

source Voc in series with an internal resistance Rint.

Both the voltage source and the internal resistance are

dependent on the state of charge (SOC) of the battery.

Based on the equivalent circuit, we have the following two

equations to describe it.

U2 ¼ Voc � Rint � I2 ð1Þ

and

P2 ¼ U2 � I2 ð2Þ

Equations (1) and (2) above are combined with Ohm’s law

to find the current, which is a function of the SOC, Voc, and

power P2 (negative during charging) to the battery. Two

solutions are mathematically possible, but only the smaller

one is physical because battery terminal voltage is limited

to a certain range around that of the open circuit [10].

I2ðSOC;P2Þ

¼VocðSOCÞ �

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ðVocðSOCÞÞ2 � 4 � RintðSOCÞ � P2

q

2RintðSOCÞ

ð3Þ

In order to study the optimization charging planning for EV

battery, a battery cell is given in this paper. The battery

parameters depend on the information of certain cell

characteristics and the size of the battery pack, which are

shown in Fig. 1 and Table 1, including the datasheet

parameters for the Saft VL 45E [18]. This type of model is

commonly used when researching the optimal control of

EVs. Eint0 and Q0 represent the maximum energy storage

and total capacity of the battery cell, respectively. Other

parameters Voc, Rint, and so on have been defined using the

model of above equations and using information on the

datasheet. The equivalent resistance Rint is approximately

constant for the majority of the charge cycle. Therefore, in

this paper, it is modeled as a constant resistance.

The battery pack for electric vehicle is scaled according

to the scaling equations in Table 1. Note that the scaling

variables ns and np are not necessarily integers. Conse-

quently, the resulting battery pack after scaling may not be

implementable in a vehicle. It is however assumed that the

behavior of the pack reflects the real behavior sufficiently

well.

3 System architecture and assumptions

In general, two kinds of control architecture can be

deployed for the optimal charging of EVs, one is called

centralized control and the another is named decentralized

0 0.2 0.4 0.6 0.8 12.5

3

3.5

4

SOC x[−]

Voc

[V]

0 0.2 0.4 0.6 0.8 11

2

3

4

SOC x[−]

Rin

t [m

Ohm

]

Fig. 1 The open circuit voltage and internal resistance of the Saft VL

45E cell

Neural Comput & Applic

123

control, the main difference lies in the deployment of the

controller in different position. For the centralized control,

the controller is put on the aggregator level, and for the

decentralized control, the controller is located at the indi-

vidual EV level. Both of the centralized and decentralized

controls have its advantages and disadvantages. For cen-

tralized control, the aggregator can aggregate large popu-

lation of EVs and then have more competence in the

electricity market, for example, may have the chance to

buy cheaper electricity, provide ancillary service to grid

more stable. However, the system operator would require

significant communication ability with EVs and powerful

computation capability. While decentralized control can

release the highly requirement on communication, but the

drawback is that the individual EV needs to collect and

store trip history and that, if EVs need to consider their

charging schedule with grid constraints, the need for

communication will be high.

Consider the needs of an optimal control study, cen-

tralized control architecture is presumed, in which a single

entity (aggregator) directly controls the charging strategy

of every vehicle to facilitate smart charging [11], and each

vehicle indirectly accesses to electricity market through

this aggregator, which is a smart interface between EV

fleets and market to play a role of coordinating charge and

discharging operation of multiple vehicles. With this cen-

tralized control, there exists an underlying assumption, that

is, the existence of contracts between aggregator and

consumers, which enable the aggregator send explicit

control signal to charge or discharge the EVs. The general

information flow is depicted below:

Figure 2 shows that aggregator is fed with following

data for charging plan making: predicted electricity price,

future driving pattern, grid constraints, and EV data, such

as EV model and state of charge of EV battery. If Evs

implemented in a large scale, peak load increases signifi-

cantly and grid may be destabilized. In this case, grid

constraints are essential for aggregator. In our study,

however, only one vehicle’s charging schedule is resear-

ched, which means grid constraints could be neglect. With

all of these information, the aggregator can make an opti-

mal control strategy for EV.

In order to achieve these centralized control, some

assumptions are given and listed below:

• The aggregator is set up to be a price taker, which

means the aggregator does not have a sufficiently large

market share to affect electricity price.

• The electricity price is assumed to be known by the

aggregator, while in the reality, the aggregator needs to

predict the electricity spot price.

• An automated communication technology is exist to

enable the smart charging, that is, all information of

EVs can be immediately communicated to aggregator,

and the control signal generated by the aggregator can

be delivered to EVs.

• In order to have a successful charging plan, a repre-

sentative driving pattern is essential. Normally, intra

city or short-term driving patterns are largely predict-

able due to fixed working hours and fixed business

schedules and routes. Therefore, a future driving

pattern is assumed to be obtained by estimating data

of past trips or established driving plans. Moreover,

electricity demand of every trip is also needed to be

assumed based upon driving pattern.

4 Control task formulation

The following notation will be used throughout this paper.

Since the market with day-ahead pricing is assumed, the

charging plan covers an entire day. For this short-term

planning, the time horizon [0, N] of a day is discretized into

equidistant time intervals [k, k ? 1] with k ¼ 0; . . .;N � 1:

It is assumed that the time interval is Dt:

This problem is addressed by considering the following

discrete system which describes the battery:

xkþ1 ¼ Tðxk; uk; kÞ ð4Þ

State variable xk represents the state of charge (SOC) of the

battery at time k. xk is not only discrete in time (index k)

but also in value. Any value has to be included in the

predefined set X, which can be calculated by a function of

charge Qk and total capacity Qmax.

xk ¼Qk

Qmax

ð5Þ

uk in Eq. (4) is the control variable, which is dimensionless

and discrete. Pk is the charge power when plug-in. In order

to obtain Pk, uk is multiplied with the maximum available

charge power (Pmax-plug) when plug-in. The electric vehicle

discussed in this paper is purely electric propulsion system,

which is characterized by an electric energy conversion

Table 1 Parameters of the VL 45E cell

Variable VL 45E Unit Scaling

Eint0 590.4 kJ Eint0 � np � ns

Q0 45 Ah Q0 � np

Vmax 4 V Vmax � ns

Vmin 2.7 V Vmin � ns

Imax 100 A Imax � np

Voc See Fig. 1 V Voc � ns

Rint See Fig. 1 ohm Rint � ns

np


123

chain upstream of the drive train, roughly consisting of a

battery (or another electricity storage system) and an

electric motor with its controller [12]. EV does not have an

internal combustion engine to provide power for

propulsion. Battery must be charged from an external

electric network. Due to this fact, the values of uk are fixed

at 0 when driving, while these values range from 0 to 1

when plug-in. If Uplug is set that covers all possible values

of uk, its discretization may be described as follows:

uk ¼uk 2 Uplug; k 2 Kplug

uk ¼ 0; k 2 Kdriv

�

ð6Þ

Kplug is a set of indices k within the time periods when the

vehicle is plugged in, while Kdriv refers to the driving

intervals. The summation of the number of elements in

Kplug and Kdriv is N, which denotes the total number of time

intervals. Any index k in Kplug or Kdriv has to be element of

the predefined set K.

k 2 K ¼ fKplug;Kdrivg ð7Þ

A specific control strategy is denoted by

u ¼ fu0; u1; u2; . . .; uN�1g ð8Þ

Any value of uk has to be element of a predefined set U, which

known as set of admissible decision. The total cost of a

sequence, f0U, is given by the cost of the final step, fN(xN), plus

the cost for all other steps, vk(xk, uk, k), then we have:

f U0 ðx0Þ ¼ fNðxNÞ þ RN�1

k¼1 vkðxk; uk; kÞ ð9Þ

To minimize the objective function (4.9), the optimal

control strategy u� ¼ fu�0; u�1; u�2; . . .; u�N�1g has to be

obtained. This is a classic dynamic programming

formulation and can be solved as described by the

literature [13, 14, 19]. The optimal trajectory is

calculated starting with the cost of the last step and going

backwards through time until the first state’s optimal cost

f0*(x0) is given by the algorithm. The recursive equation is

listed as follows:

fkðxk; ukÞ ¼ minfvkðxk; ukÞ þ fkþ1ðxkþ1Þg¼ minfvkðxk; ukÞ þ fkþ1ðTðxk; uk; kÞÞg

ð10Þ

u�k ¼ argminðfkðxk; ukÞÞ ð11Þ

According to dynamic programming, some special terms

are given as follows:

k: step

X: set of admissible state

U: set of admissible decision

Admissible state set X must be defined appropriately,

because T(xk, uk, k) may not be any of the elements of X

where fk?1 is know if set X is not defined good enough,

which means T(xk, uk, k) may not equal with xk?1 as

described in Eq. (4). Actually in practice operation, errors

are hardly avoided that T(xk, uk, k) will usually not be on a

grid point no matter how the set X is defined. Therefore, an

approximation is needed. Normally, xk?1 is defined as a

range by plus a margin of 10 % to cover the possible

T(xk, uk, k).

4.1 Cost of final step

The objective function is to ensure that the battery is fully

charged before the first trip of the following morning. Step

N is the moment right before the next day’s departure.

Therefore, the value of xN is 100 %, and no charging

operation happens at step N. The cost of final step fN(xN)

should be defined as.

fNðxNÞ ¼ 0 ð12Þ

4.2 Cost of other step

For EV, a purely electric propulsion system, it is important

to acknowledge different step function vk for driving mode

and plug-in charging mode. The most general case is given

by introducing the following discontinuity:

Fig. 2 Information flow in

centralized control


123

vkðxk; uk; kÞ ¼vplugðxk; uk; kÞ; k 2 Kplug

vdrivðkÞ; k 2 Kdriv

�

ð13Þ

where

vplugðxk; uk; kÞ ¼ gk � uk � Pmax�plug � CelðkÞ � Dt ð14Þ

and

vdrivðkÞ ¼ 0 ð15Þ

gk denotes the efficiency parameter between step k and

k ? 1. Cel is the price of electricity per unit of energy. Dt is

the time interval between step k and k ? 1. Since EV is a

purely electric propulsion system, charging cannot take

place during driving period as hybrid electric vehicle did,

and the charging cost is set to 0.

4.3 Equation of state transition

In order to well describe battery charging, a first-order

system is developed with Backward Euler method applied

to Eq. (5), where xk is an actual SOC at the step k, xk?1 is a

desired SOC at step k ? 1.

xkþ1 ¼ Tðxk; uk; kÞ ¼ xk þ Dt � _xkþ1

¼ xk þ Dt �_Qk

Qmax

¼ xk þ Dt � I2ðxk; kÞQmax

where

I2ðxk; kÞ ¼VocðxkÞ �

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ðVocðxkÞÞ2 � 4 � Rint � PBTðkÞns

q

2Rint

ð16Þ

that comes from Eq. (3) given in Sect. 2. The battery

current is calculated using

PBT ¼ ns � U2 � I2 ð17Þ

and

PBTðkÞ ¼�gk � uk � Pmax�plug; k 2 Kplug

Pdr; k 2 Kdriv

�

ð18Þ

Equations (17) and (18) are deduced based on Sect. 2 EVs

are considered to be battery packs in this study, which is

implied with a parallel wiring, resulting in an equal power

contribution of all cells. Pdr(k) denotes power requirement

during driving cycle. It is obvious that power requirement

of every step when driving is hardly predicted. Instead, it is

possible to predict energy requirement throughout a whole

driving trip, which can be a replacement of power

requirement. Note that the relation shown in Eq. (17) is

only exact for infinitesimal time intervals. To achieve

sufficient accuracy, time interval choosing is very important.

It should be done in a way that maximum charge Dt � I2

is small enough and does not influence the results of the

battery equations substantially.

These state transition models [Eq. (16)] are general

enough to account for the possibility of parallel processing

among the various control strategies, as well as for

redundancy in the database. Once the concept of state

transition has been properly defined, dynamic program-

ming can be used to find the state containing the answer to

the query that has the minimum cost and to find the optimal

trajectory to that state (i.e., optimal sequence of processing

operations) [15].

5 Case study

In this section, a case is studied and the goal of optimi-

zation is to present a charging schedule for every individual

vehicle to minimize the cost of electricity while satisfying

the vehicle owner’s requirements. In this case, vehicle

would be plugged in every time when the driving finished.

A comparison is made between the results of an EV with a

fast charging scheme and those of the dynamic program-

ming-based method.

The charging schedule is divided into time intervals for

a 24-h-based period. The period starts from the first second

when the vehicle owners begin their first trip and ends right

before the next day’s departure, which has 288 intervals of

5-min each. As mentioned previously, the objective func-

tion is to ensure that the battery is fully charged before the

first trip of the following morning, which means

x0 = 100 % and xN = 100 %. The vehicle used to study in

this paper is a battery packs, which is a purely electricity

propulsion system. The basic battery information is shown

in Sect. 2, and more additional is listed in Table 2. U is

defined as the set of admissible state which indicate the

possible control signals and contains 11 elements which

limited in ½0; 0:1; 0:2; . . .; 1:0�: X is defined as the set of

admissible decision which indicate the possible SOC of the

battery and contains 101 elements which limited in

½0; 1; 2; . . .; 99; 100 %�. Dt is time interval, which is defined

as 5 minutes. Here, we give ns = 100 and np = 1.12.

Typically, the battery operation is limited to a given state

of charge operating range. It is assume that battery has a

minimum state of charge of 10 %. Hence, here the SOC is

limited between 10 and 100 %. Moreover, it is important to

know a driving behavior of a vehicle, which includes the

departure time, return time, and energy requirements of

every trip. Based on the vehicle parameters, a driving map

includes three trips during a day is given in Table 3.

Another important piece of information is electricity

prices, which are based upon a typical work day of

04.05.2011 from Nordpool Spot market area Denmark

West [17]. To obtain an optimal charging schedule, a price


123

for electricity service is prerequisite. Any optimization

should take use of real prices of one day.

5.1 Fast charging

Fast charging is a kind of uncoordinated charging. It

assumes that vehicles owners face a flat price throughout

the whole and consequently their vehicles are charged

instantaneously when they are plugged in, and the batteries

will be fully recharged as fast as possible without consid-

ering the daily electricity price. Some vehicle manufac-

turers, such as Nissan, give their own definition for battery

fast charging. For example, Nissan LEAF’s battery is

intended to accept several rapid charging scenarios

including a 50 KW ‘‘fast charge’’ which gives 80 % charge

in thirty minutes, or a five-minute fast charge which

delivers an additional 31 miles of range. These rapid

recharge modes will require a special three-phase charger,

which is most likely to be owned by commercial or gov-

ernmental entities in distributed charging stations [16].

This fast charging offers most flexibility to driver. How-

ever, it is not the fast charging we discussed here, because

homeowners do not have a spare 50 KW charging power,

but prefer to have a common, single-phase 220 V with

maximum 4 KW charging power. The profiles with fast

charging are given by Fig. 3. It is obvious that every time

when the vehicle finishes a trip, the battery will be charged

immediately without considering the electricity prices. The

battery is fully charged as fast as possible once it is plugged

in. The fast charging strategy provides customers with

flexibility; however, the electricity costs will be high,

which for the profiles amount to EU 2.2415.

5.2 Smart charging

The idea of the smart charging is to achieve optimal

charging to minimize the charging cost. The results of the

fast charging algorithm is compared to an EV with a

dynamic programming-based smart charging scheme. With

the automated communication technology, all information

can be immediately communicated to aggregator, which

then returns a charging plan for an individual EV for the

following day. The optimal control strategy will be

obtained and sent to the individual vehicle as control sig-

nals for charging power. It is expected that most of

charging occurs when given the comparably lower prices.

The results of the dynamic programming-based method are

shown in Fig. 4. The simulation parameters are given in

Table 2.

From Fig. 4, we have a general idea that electric vehicle

charging is done when the price for electricity is lowest.

The SOC of battery shows that the battery does not have to

be fully charged before next trip. Instead, it would be

sufficient if the SOC is charged enough to support the

energy consumption for the next trip. This leads to a

electricity cost of EU 1.8333 for a whole day, which is

cheaper than the fast charging cost. Smart charging cannot

offer flexibility for driver as fast charging does. Conse-

quently, sometimes when drivers drive away their vehicles

before the preannounced departure time, the battery may

not be enough charged to meet the energy requirement for

the next trip, which the drivers have to accept. It is also

interesting that when the vehicle is charged, it always does

Table 2 Simulation parameters

Discretization Parameters

U 11

X 101

Dt 300 s

Battery Value

Total capacity 50 Ah

Maximum energy storage 18.4 KWh

Maximum plug power 4 KW

Rint 0.0025 ohm

Table 3 Driving behavior

Trip Departure time Return time Energy

requirement (KWh)

1 8:00 9:00 11.04

2 15:00 16:00 13.25

3 20:00 21:00 11.04

0 5 10 15 20 250.05

0.06

0.07

0.08

13:008:00 18:00 23:00 4:00 8:00

Time (h)

13:008:00 18:00 23:00 4:00 8:00

Time (h)

13:008:00 18:00 23:00 4:00 8:00

Time (h)

Ele

ctric

ity P

rice

0 5 10 15 20 250

0.5

1

1.5

Con

trol

Str

ateg

y u[

−]

0 5 10 15 20 250

0.5

1

1.5

Sta

te o

f Cha

rge

x[−

](e

u/K

Wh)

Fig. 3 Profile with fast charging


123

its charging with the maximum available power, which

means uk is equal to 1 when xk is increasing. This is caused

by the fact that intraday price differences are higher than

loss costs and supported by increased inverter efficiency at

high power throughputs [10]. Due to this consideration,

some paper studies on control of charing sequence with

charging rate fixed to the maximum, instead of charging

rate control, such as [3].

5.3 Comparison for the simulation results

The code for smart charging optimizes a 24-h interval in

Matlab takes 13.19 s on 2.1-GHz CPU with 1.99 GB of

RAM.

The following Table 4 gives a comparison between two

charging algorithms. Each has both advantages and dis-

advantages as can be seen from the table. However, it is

widely recognized that electric vehicle will be a control-

lable load in the near future, which is more intelligent and

has more communication ability, even though giving a

smart charging schedule takes more time due to compli-

cated charging control algorithm.

6 Conclusions

This paper presents a mathematical formulation and a

dynamic programming-based algorithm for optimizing an

EV’s charging schedule with given electricity prices and

driving pattern within a centralized control architecture.

Smart charging without provision of regulation service

reduces daily electricity costs for driving from EU 2.2415

to EU 1.8333 compare with fast charging. With smart

charging, EV is recharged during the lowest electricity

price period, where is also the off-peak hours. It naturally

drops the possibility of grid overload during the peak load

hours.

More work could be done in the future. Only one

electric vehicle’s charging schedule has been researched

here. Studies of optimal control on large number of EVs

should be done, which involves high requirement on

communication, and possibility that EV charging may

impact the electricity price. Therefore, decentralized con-

trol architecture can be considered and electricity price

forecasting models should be properly developed. Fur-

thermore, the optimization model should be extended to

account for providing of regulation service and given dif-

ferent type of electric-drive vehicles as well as various

driving patterns.

Acknowledgments This work was supported in part by the National

Natural Science Foundation of China (grant no. 61005090, 61034004,

61075064, 61003053), the Program for New Century Excellent Tal-

ents in University of Ministry of Education of China (grant no.

NECT-10-0633), the PhD Programs Foundation of Ministry of Edu-

cation of China (grant no. 20100072110038), the International S&T

Cooperation Program from Ministry of Science and Technology of

China (grant no.2011DFG13020), the Shanxi Province Natural Sci-

ence Foundation of China (grant no. 2011011012-1), and the Program

for the Top Young Academic Leaders of Higher Learning Institutions

of Shanxi.

References

1. Han S, Han S, Sezaki K (2010) Design of an optimal aggregator

for vehicle-to-grid regulation service. Proc Innov Smart Grid

Technol (ISGT) 19–21:1–8

0 5 10 15 20 250.05

0.06

0.07

0.08

Ele

ctric

ity P

rice

0 5 10 15 20 250

0.5

1

1.5

Con

trol

Str

ateg

y u[

−]

0 5 10 15 20 250

0.5

1

1.5

13:00 18:00 23:00 4:008:00 8:00

Time (h)

Sta

te o

f Cha

rge

x[−

]

13:00 18:00 23:00 4:008:00 8:00

Time (h)

13:00 18:00 23:00 4:008:00 8:00

Time (h)

(eu/

KW

h)

Fig. 4 Profile with smart charging

Table 4 Comparison between fast and smart charging

Charging algorithms Advantages

Fast charging More flexibility,

Fast computing time

Smart charging Low charging cost,

More intelligent, controllable,

More communication ability

Charging algorithms Disadvantages

Fast charging High charging cost,

Less intelligent, uncontrollable,

Less communication ability

Smart charging Less flexibility

Comparably slow computing time


123

2. Kempton W, Tomic J (2005) Vehicle-to-grid power fundamen-

tals: calculating capacity and net revenue. J Power Sources

144(1):280–294

3. Han S, Han SH, Sezaki K (2010) Development of an optimal

vehicle-to-grid aggregator for frequency regulation. IEEE Trans

Smart Grid 1(1):65–72

4. AC Propulsion (2005) Development and evaluation of a plug-in

hev with vehicle-to-grid power flow AC propulsion, CARB Grant

Number ICAT 01-2, 2005

5. Tomic J, Kempton W (2007) Using fleets of electric-drive vehi-

cles for grid support. J Power Source 168(2):459–468

6. Kempton W, Tomic J (2005) Vehicle-to-grid power implemen-

tation: from stabilizing the grid to supporting large-scale

renewable energy. J Power Sources 144(1):280–294

7. Heydt G (1983) The impact of electric vehicle deployment on

load management strategies. IEEE Trans Power Apparatus Syst,

PAS-102 5:1253–1259

8. Clement-Nyns K, Haesen E, Driesen J (2010) The impact of

charging plug-in hybrid electric vehicles on a residential distri-

bution grid, IEEE Trans Power Syst 25(1): 371–380

9. Sundstroem O, Binding C (2010) Planning electric-drive vehicle

charging under constrained grid conditions. In: International

conference on power system technology, Hangzhou, China,

pp 1–6

10. Rotering N, Ilic M (2011) Optimal charge control of plug-in

hybrid electric vehicles in deregulated electricity markets. IEEE

Trans Power Syst 26(3):1021–1029

11. You S, Traehold C, Poulsen B (2009) A market-based virtual

power plant. In: Proceedings of the international conference clean

electrical power (ICCEP) 460–465

12. Guzzella L, Sciarretta A (2007) Vehicle propulsion systems:

introduction to modelling and optimization. ETH Zuerich,

Switzerland

13. Bellman R (1957) Dynamic programming. Princeton University,

Princeton

14. Bertsekas D (2007) Dynamic programming and optimal control.

Athena Scientific, Nashua

15. Lafortune S, Wong E (1986) A state transition model for dis-

tributed query processing. ACM Trans Database Syst 11:294–322

16. Nissan battery documents, http://www.nissanusa.com/leaf-electric-

car/tags/show/battery#/leaf-electric-car/tags/show/battery

17. http://www.nordpoolspot.com/

18. Saft Batteries, High energy lithium-ion cell VL 45E cell, Jan

2010, [Online] http://www.saftbatteries.com

19. Kang Qi, Wang Lei, Wu Qidi (2009) Swarm-based approximate

dynamic optimization process for discrete particle swarm opti-

mization system. Int J Bioinspired Comput 1(1/2):61–70


123

Numerical comparison of optimal charging schemes for electric vehicles 119

A.4 Numerical comparison of optimal chargingschemes for electric vehicles

This paper was presented at IEEE PES General Meeting, San Diego, U.S., July2012.

1

Abstract-- The optimal charging schemes for Electric vehicles (EV) generally differ from each other in the choice of charging periods and the possibility of performing vehicle-to-grid (V2G), and have different impacts on EV economics. Regarding these variations, this paper presents a numerical comparison of four different charging schemes, namely night charging, night charging with V2G, 24 hour charging and 24 hour charging with V2G, on the basis of real driving data and electricity price of Denmark in 2003. For all schemes, optimal charging plans with 5 minute resolution are derived through the solving of a mixed integer programming problem which aims to minimize the charging cost and meanwhile takes into account the users' driving needs and the practical limitations of the EV battery. In the post processing stage, the rainflow counting algorithm is implemented to assess the lifetime usage of a lithium-ion EV battery for the four charging schemes. The night charging scheme is found to be the cheapest solution after conducting an annual cost comparison.

Index Terms—Electric vehicle, mixed integer programming, optimal charging, rainflow counting, vehicle-to-grid, V2G

I. INTRODUCTION HE technique of optimal charging in the context of deregulated electricity markets, sometimes referred as

smart charging, has recently caught a lot of attention, as there is a tremendously growing need for electrifying the transportation sector [1]. The term “optimal” can be generally interpreted from two perspectives: the EV owners and the power system operators.

From the EV owners’ point of view, “optimal” can be simply interpreted as minimizing the cost of charging while guaranteeing their need for driving. In [2]-[5], this perspective has been intensively investigated based on various modeling techniques including linear programming, dynamic programming and quadratic programming. A common impression inferred by these studies is that the optimal charging as a feasibility solution can considerably reduce the cost of charging; however, to support a large-scale roll-out of EV, services like V2G have to be offered by EV or EV fleet to improve the EV economy [6]-[7].

From the power system operators’ point of view, “optimal” can be generally interpreted as a complex objective which aims to maximize the advantages of EV and minimizes its

This work is part of project ‘EcoGrid EU’ with EU funding. The authors are all with the Centre for Electric Technology, Department of

Electrical Engineering, Technical University of Denmark, Kgs. Lyngby DK2880, Denmark (e-mail:{sy, junhu, abp, pba, cnras, stc} @elektro.dtu.dk).

disadvantages. The energy storage nature of EV makes it a potential solution to many power system problems, such as load shifting and frequency regulation. Meanwhile, its nature of being a mobile electrical load challenges the power system operation, as an inappropriate integration could easily cause voltage issues and overloading in the distribution network. Regarding this aspect, intensive studies have been done in [8]-[11], wherein coordinated charging schemes for an EV fleet with either centralized or decentralized control structures are developed to handle grid constraints and meet the driving requirements at the same time.

Apart from the different perspectives of the two groups of stakeholders, the optimal charging schemes are also heavily dependent on a large number of factors, including:

1) range of uncertainty related to electricity price anddriving pattern [12];

2) fidelity of the EV battery models which varies fromlinear to non-linear;

3) modeling approaches used to describe the chargingprocess and the associated optimization techniques which span from the conventional linear programming to the genetic algorithms [13];

4) time resolution used in the various simulations thatspans from a few minutes up to hours;

and so on. In this study, four different optimal charging schemes for

EV, namely night charging, night charging with V2G, 24 hour charging and 24 hour charging with V2G, are formulated as a set of mixed integer programming problems. Based on the practical driving data and electricity prices collected for Denmark in 2003, these charging schemes are simulated at a time resolution of 5 minute. Objectives for the four charging schemes are set to charging cost minimization, due to the fact that today’s EV penetration is relatively low compared to the other Distributed Energy Resources (DER) technologies and some distribution grids can handle the penetration level up to 20% [14]. In the following text, the mathematical formulation for various optimal charging schemes is described in Section II. Section III presents the numerical model development.Results and discussions are summarized in Section IV, wherein battery life and annual cost comparison among the four schemes are included. Section V concludes the study.

II. PROBLEM FORMULATION

Today, energy procurement of EVs into the power

Numerical Comparison of Optimal Charging Schemes for Electric Vehicles

S. You, Member, IEEE, J. Hu, Student, IEEE, A.B. Pedersen, Student, IEEE, P.B. Andersen, Student, IEEE, C. N. Rasmussen and S. Cha

T

2

system is very market-driven, given the fact that the economy of EVs can be improved by market participation via appropriate aggregation services, e.g. Virtual Power Plants (VPP). A comprehensive analysis of different aggregation setups has been carried out by the Danish EDISON project in the last few years [15]. The report from the EDISON project suggests three methods that can facilitate the EVs' participation in today's electricity spot market:

1) The retailer broadcasts the electricity price once a day to the individual EV owners and the EV owners therefore make appropriate charging schedules according to the known price and the local intelligence;

2) The charging strategy can be a simple time of day charging based on EV owners' empirical knowledge on when the electricity price is relatively cheap, such as at night;

3) THE charging of EVs can be scheduled or controlled by a fleet operator based on a contractual setup, where the central intelligence is more relied upon.

In this study, the first integration method is utilized and modified to accommodate four different charging schemes:

1) Night charging: the charging period is constrained to be from 7pm to 7am, and discharging is not allowed;

2) Night charging with V2G: the charging or discharging actions can be performed between 7pm to 7am;

3) 24 hour charging: the charging can be performed anytime when the EV is not in use, and discharging is not allowed;

4) 24 hour charging with V2G: the charging or discharging actions can be performed anytime in a day.

Since the market segment of today's EVs is primarily urban area due to the battery capacity limitation and public health concerns, this study firstly assumes that the charging infrastructure is available everywhere in the studied urban area, indicating that the EVs can be charged or discharged as long as they are not used for driving. Further, the terminology “V2G” used in this study refers to selling the battery energy back to the grid on an hourly basis at prices set the previous day; while in other literature V2G is normally referred to a mechanism that activates the provision of ancillary services.

To enable the comparison study, a mixed integer programming formulation is developed and solved using Matlab to find the optimal solution for each charging scheme applied to an individual EV. The common objective of each charging scheme, as in (1), is to minimize the cost of charging given the broadcasted electricity price and energy required for driving ; meanwhile, the four charging schemes are enabled separately by turning on/off the binary variables and .

∆ ∙ ∙ +∆ ∙ ∙ ∙

(1) s. t.

= + {∆ ∙ + ∆ ∙ − ∙ }∙ ≤ ≤ ∙ + 1 ∙ + 1 ≤ 0 ≤ ∆ ≤ , ∙ ∙ ∆ − , ∙ ∆ ≤ ∆ ≤ 0+ + = 1 (2) where the planning duration is divided into time intervals with denotes the number of sequence and ∆ denotes the time length of each interval. Decision variables ∆ and ∆ represent the energy charged into and discharged from the battery in each time interval respectively, while the other three binary variables , , and indicate the on/off status of charging, V2G and driving for each corresponding time interval. To facilitate the expression, an intermediate variable

is introduced to represent the energy level of the battery in the end of each time interval.

Parameters and represent the nominal energy capacity and the initial energy of the battery in the planning period, while the charging and discharging efficiency are represented by and . The maximum power exchanged between the EV inverter and the electrical grid is expressed by , and , respectively during charging and discharging processes, which constrains the maximum energy exchanged between the EV and the grid. For battery life concerns,

and are further introduced to represent the manufacturer recommended sate of charge (SOC) range.

Explanations for the inequality constraints can be found in [3] wherein a similar problem is described using linear optimization by the same group of authors. Compared to the previous study, a major improvement is made in this study by formulating various charging schemes in a more generic and flexible way. Power performance of the battery is not included in this paper; however, given the energy performance within a certain time, the power performance can be derived simply by elaborating on the charging schemes e.g., constant power and constant current.

III. NUMERICAL MODEL DEVELOPMENT For this numerical case study, the optimal charging plans

for different charging schemes are derived for the next day with 5 minute resolution given the broadcasted electricity price and EV owners pre-defined driving requirement. The assumption of flawless forecasting made in this paper aims to resemble the best case scenario for different charging schemes. In practice, EV owners may not be able to precisely forecast their driving needs for the next day with 5 minute resolution in practice, which could result in higher cost for charging.

The daily optimization is further repeated over the course of a month to retrieve a more general charging performance. In this section, the selected battery model parameters, the driving information and the source of electricity hourly prices are briefly explained.

A.

ionau28baranchEVdismaandispo,ne4kassmoexroufor

beillu

Fig Thexba whco

B.

thewaveinfwh14

Battery ModeDue to its hig

n battery techutomotive indu8kWh Lithiumattery of a mednge is set to b

harging and disV is grid connscharging ,aximum powe

nd the maximuscharging. Wh

ower discharge, as the diseeds e.g., accelekW. For everysumed to be orning driving

xchanged betweund trip efficier driving in thaFor the emula

etween the lifeustrated in Fig

g. 1. Lifecycle vs.

his relationshxponential equaattery lifetime e = 4515.2 ∙here is DOD

orresponding to

Driving InforAs few electr

e related studieay that vehicleehicle ownersformation is thich 360 cars i4 to 100 days. D

el gh density on phnology has bestry for vehicle

m-ion battery dium size fambetween 10% ascharging are bnected, the ma

and , arr drawn from

um power recehen the EV is ued to drive thscharging poweration, and cay single day, 50% to guara

g. This assumpeen the batteryency) within a at day. ated Lithium-ioecycle and the .1.

DOD of a Lithium

hip is approation, as in (3estimation.

.

in %, is the o any given val

rmation ic vehicles arees of EVs, a gee electrificatio' driving pattaken from thin CopenhagenDue to the data

power and eneeen commonlye electrificatiois emulated

mily car in Denand 90%, and

both assumed aaximum powere both set to 4grid is 4.4kW

eived by grid used for drivin

he motor is noer is dependen

an therefore bethe initial S

antee the neeption also indy and the grid day is equal to

on battery, a geDepth of Dis

m-ion battery [17]

oximately ex), to support t

number of expue of .

e already on theneral assumption would havettern. In this

he 2003 AKTAn were trackeda incompletene

ergy, the Lithiuy adopted by n. In this studyto represent nmark. The Sthe efficiency

as 90%. When r of charging a4kW, meaning

W during chargis 3.6kW dur

ng, the maximot constrained nt on the driv much larger th

SOC at 00:00d for very eaicates the ene(disregarding

o the energy u

eneral relationsscharge (DOD

xpressed by the later study

pected cycles

e road, to suppion is taken in e little impact s study, drivA survey[16],

d using GPS fress, the full-mo

um-the

y, a the OC

y of the and the

ging ring

mum by

ving han

0 is arly ergy

the used

ship ) is

an y on

(3)

port the on

ving , in rom onth

record oto suppoTable 11kWh/requiredFig.2. Tthe fact family a

20

20

200

200

200

200

200

200

200

Fig. 2. En

C. ElecThe

Nordic basis bytime coinformaas illuststudy. Telectricistudy.

Withthan 25%is knowwith larcontext,also relaoperatiosuch as benefitsancillary

of the vehicle nort this study. I. Following

/100km, thed for driving The frequent dr

that the selecand shared by i

SAMPLE OF

Start 003-3-2 9:31:57

003-3-2 9:44:21

03-3-2 10:45:38

03-3-2 10:55:56

03-3-2 12:58:36

03-3-2 15:24:21

03-3-2 17:23:21

03-3-2 18:41:00

03-3-2 18:48:39

nergy required for

ctricity Price information opower market

y the market oponsistency be

ation, the electtrated by the fiThe figure onity price on M

h its average w% of its annua

wn as a good rrge-scale stoc, the volatilityatively high, won. For other

DK-east, the s due to the rouy services whic

no. 32139 for The original d

g the EV AKTA data with 5 minutriving behaviocted vehicle is its family mem

TABLE I DRIVING DATA FR

Finish D9:40:23

9:46:59

10:50:53

11:00:14

13:02:24

15:27:34

17:26:09

18:44:57

18:52:54

driving with 5 mi

of hourly elect is published perator Nord Petween electrtricity price of igure on the len the right in

March 2nd, whic

wind power pl electricity de

representation chastic reneway of the electrwhich to a grearegions with mV2G operation

und-trip energych is not consid

March 2003 idata format is imileage assuis converted

te resolution aors within a da

owned by a mmbers.

ROM 2003 AKTADuration Dis0:08:26 2

0:02:38

0:05:15

0:04:18

0:03:48

0:03:13

0:02:48

0:03:57

0:04:15

inute resolution on

ctricity spot pand recorded

Pool Spot. To gricity price af DK-west in Meft in Fig.3, is u

Fig.3 depictsch is used in th

production covemand, the regiof a future en

able resourcesricity price inat extent benefmuch less prin could hardlyy loss unless itdered in this st

3

s picked out illustrated in umption of

d to energy as shown in ay are due to medium size

tance(m) 21278

1685

2855

2336

2206

2110

1658

2587

2689

n March 2nd

price of the d on a daily guarantee the and driving March 2003, used for this s the hourly he later case

vering more ion DK-west

nergy system . In such a DK-west is

fits the V2G ice volatility y incur extra t is traded as tudy.

Figthe

beancosch

A.

byreqchschthasch3ppeschpeele

Fig

sch

g. 3. Hourly electe left), and on the d

IVIn this sectio

ehavior of the fnd the short onsumption as heme are calcu

Intra-day PerThe result of

y Fig. 4, whquirement for d

harged (positivehemes. Amongat the hours wihemes; howev

pm, for the 2erformed in thhemes with V

eriods become ectricity arbitra

g. 4. Charging pro

The correspohemes are give

tricity price of DKday March 2nd (to t

V. RESULTS AN

on, the intra-dfour different c

term economwell as the an

ulated and com

rformance an intra-day st

here the bluedriving and thee)/discharged (g the chargingith cheap electer as the lowes24 hour charghat period instV2G, charging

much more frage.

ofiles with 5 minut

nding SOC vaen in Fig. 5.

K-west in the monthe right)

ND DISCUSSION

day and the mcharging schemmic performannnualized cost fmpared.

tudy for Marche line represe red lines repr(negative) for d

profiles, it caricity are mostst electricity prging scheme, ead of midnigand dischargin

frequent in ord

te resolution on Ma

ariations for d

nth of March 2003

NS monthly charg

mes are illustratnce, the lifetifor each charg

h 2nd, is illustraents the eneresents the enedifferent charg

an easily be foutly selected forrice occurs arou

the chargingght. For chargng in non-drivder to profit fr

arch 2nd 2003

different charg

3 (to

ging ted, ime

ging

ated ergy ergy ging und r all und

g is ging ving rom

ging

Fig. 5. SO The r

on Marccase, thmuch menrichedchargingelectricielectrici

nightnight24h c24h c

B. MonBy re

and thepresenteV2G” isthird ofthere is“24h chare mos

nightnight24h c24h c

OC profiles with 5

resulting chargch 2nd 2003 a

he charging schmore cost-effed flexibility; wg and 24h chity price in theity price at nigh

CHARGING COS

Charging Optt charging t charging + V2Gcharging charging + V2G

nthly Performaepeating the ine charging coed in Table IIs the least exp

f the most exps little cost diharging” due tstly in midnigh

CHARGING CCharging Opt

t charging t charging + V2Gcharging charging + V2G

5 minute resolution

ging costs for dare listed in Tahemes with Vective than thwhile the diffharging is vere day time is vht.

TABLE II ST COMPARISON FO

tions

ance ntra-day simulaosts can be dII, the chargingpensive, whichpensive optionfference betw

to the fact thatht.

TABLE III COST COMPARISON

tions

n on March 2nd 200

different chargable II. For t

V2G options arhe other twoference betweery small as tvery close to t

OR MARCH 2ND 20Charging Cost (

0.227 -3.17850.1816-6.3705

ation, the chargderived rather g option “24hh costs approxn “night chargieen “night cht cheap electri

N FOR MARCH 2003Charging Cost (

33.476216.136633.377210.7202

4

03

ging schemes the intra-day re obviously due to the

en the night the cheapest the cheapest

003 (DKK)

ging profiles simply. As

h charging + ximately one ing”. Again,

harging” and icity periods

3 (DKK)

AmcosimthedeIn raiwhchpro90pafrobebyins90indnucoDOpro

Fi

chlifca

mong the vonsumption of mply counts the operational p

etermines the fFigure 6, an

inflow countinhich have bee

hange of SOC oocess, if one c

0° clockwise, tagoda roof. If om each peak elonging to a spy the colored stance, the rain

0%, implying adicates a half

umber of partiounted into binOD being the ocess.

igure 6: Illustratio By applying t

harging profilefetime consumlculated and ill

LIF

Charging Opnight charging night charging +24h charging 24h charging +

various ways f a battery, thhe number of cperiod and assufraction of the example that i

ng is applied is en randomly cover a certain tould imagine tthe shape of thf a raindrop is

and valley opecific raindroplines in the m

ndrop that folloa half cycle of 9

cycle of 60%ial cycles witns of equal wi

granularity le

n of the rainflow c

this post proces, while takin

mption for dilustrated in Fig

TABFE TIME ESTIMATI

ptions Life U0

+ V2G 00

V2G 0

of assessinhe rainflow cocycles for eachumes that the a

lifetime that iillustrates howpresented. Th

created, represime period. Duthat the originahe curve woulfurther assumf the roof, thep could be obtamiddle figure ows the blue pa90% DOD; wh

% DOD. In thh different vaidth, with the evel considered

counting algorithm

essing algorithmng into accounfferent chargig. 7 and Table

LE IV ION FOR EV BATTE

Usage (%) Exp0.5526 0.6899 0.5886 0.9540

ng the lifetiounting algorith level of DODamplitude of DOis consumed [1

w the algorithme original sign

sent the observuring the countal curve is rotald be similar ted to start falle half cycle pained as indicain Figure 6.

ath has a lengthhile the green phe histogram, alues of DODbin width of

d in the count

m

m to the montnt (3), the batting schemes IV.

ERY pected Life (year)

15.08 12.8 14.16 8.74

ime thm

D in OD 18].

m of nals, ved ting ated to a ling path ated For h of path

the D is

1% ting

thly tery are

Fig. 7. Li

As exsmall amoptions chargingcheapesthis schbattery “24 chathe expecompare

C. AnnuTo p

approaccost

where and therespectivdiffereneach schin Tablroughly Lithiumderived Table Vscheme the “24hsolutionthe high

ifecycle consumpt

xpected, the nmplitudes are are included

g option “24 st among the foheme turns oulifetime has b

arging + V2G”ected battery led to the night

ual Cost Compprovide an infch is introduce

for different= + and r

e charging cosvely, and

nt charging schheme is assumee III, by furth

y represents thm-ion battery, t

for each chargV. Among t

is found as thh charging + V

n due to its sevh battery capita

tion for the emulat

numbers of condramatically as illustrated

charging + Vour schemes, thut to be the labeen consumed” in the studiedlife is thereforcharging sche

parison formative overed by (4) to rot charging sche⁄

represents the st incurred du

indicates thhemes. If the med to be the saher setting he present mthe annual co

ging scheme. Tthe four optiohe most cost-eV2G” is foundvere impact onal cost.

ted EV battery in M

nsumed partialincreased whe

d in Fig.6. AV2G” is foundhe associated liargest. 0.95% d for the chargd period Marcre reduced by

eme.

rall comparisooughly estimateemes.

capital cost ofuring the batthe expected monthly charg

ame as the valu to 120,000D

market price ost for driving

The result is sumons, the “nigheffective solut

d to be the mon battery servi

5

March 2003

l cycles with en the V2G

Although the d to be the ife usage for of the total

ging scheme ch 2003, and

almost half

on, a simple e the annual

(4)

f the battery tery lifetime lifetime for

ging cost for ues presented DKK which f a 28kWh EV can be

mmarized in ht charging” tion whereas st expensive ice time and

6

TABLE V ANNUAL COST COMPARISON FOR DIFFERENT CHARGING SCHEMES

Charging Options (DKK/year) night charging 8359 night charging + V2G 9569 24h charging 8875 24h charging + V2G 13859

V. CONCLUSION This paper has provided a numerical comparison of four

different optimal charging schemes, namely night charging, night charging with V2G, 24 hour charging and 24 hour charging with V2G, on the basis of real driving data and electricity price of Denmark in March 2003. Based on a best case scenario i.e. flawless forecasting, optimal charging schemes with 5 minute resolution are found by solving a mixed integer programming problem, and compared from both short-term and long-term perspective.

It has been found that the “night charging” scheme exhibits the lowest annual cost of using the EV. On the contrary, although the V2G option can to a great extent reduce the charging cost, its severe impact on the battery lifetime noticeably increases the annual cost of using the EV. This also implies the importance of judicious designs for V2G operations, which shall not only account for the value of V2G but also suppress its negative effects. Difference between “night charging” and “24h charging” appears to be small as the electricity price is normally cheap around midnight. This may indicate the need for establishing a residential charging infrastructure could come before the need for having a public charging infrastructure.

The EV economics are heavily dependent on the pattern of electricity prices and EV users as well as the other important factors, such as tax and subsidies. The study performed in this paper is therefore more informative than conclusive. To deliver an unbiased assessment of different charging schemes, collecting the sufficient representative information, refining the modeling assumptions, investigating the grid impacts and designing appropriate validation approaches are considered as future work. For the V2G possibilities, instead of providing bulk energy back to the grid as presented in this study, it would also be interesting to examine the economy of providing ancillary services to the power system operators with various risk-averse algorithms.

VI. REFERENCES [1] J. Lopes, F. Soares and P. Almeida, "Integration of electric vehicles in

the electric power system," in Proc. of the IEEE, vol. 99, issue 1, pp. 168-83, Jan. 2011.

[2] T. Lan, J. Hu, G. Wu and S. You, “Optimal charge control of electric vehicles in electricity markets,” presented at the 4th Int. Workshop on Intelligent Information Management Systems and Technology, Shanghai, China, 2011

[3] J. Hu, S. You, and J. Østergaard, "Optimal charging schedule of an electric vehicle fleet," presented at the 46th Int. Universities Power Engineering Conf., Soest, Germany, 2011.

[4] N. Rotering and M. Ilic, "Optimal charge control of plug-in hybrid electric vehicles in deregulated electricity markets," IEEE Trans. Power System, vol. 26, issue 3, pp. 1021-1029, Aug. 2011.

[5] T. K. Kristoffersen, K. Capion and P. Meibom, “Optimal charging of electric drive vehicles in a market environment,” Applied Energy, vol. 88, issue 5, pp. 1940-1948, Dec. 2010.

[6] W. Kempton and J. Tomic, “Vehicle-to-grid power fundamentals: Calculating capacity and net revenue,” Journal of Power Sources, vol. 99, pp. 168-183, Dec. 2005.

[7] W. Kempton and J. Tomic, “Vehicle-to-grid power implementation: From stabilizing the grid to supporting large-scale renewable energy,” Journal of Power Sources, vol. 144, issue 1, pp. 280–294, Jun. 2005.

[8] O. Sundstrom and C. Binding, “Optimization methods to plan the charging of electric vehicle fleets,” in Proc. 2010 ACEEE Int. Conf. on Control, Communication, and Power Engineering, pp. 323-328

[9] J. A. p. Lopes, F. J. Soares and P. M. Almeida, “Smart charging strategies for electric vehicles: enhancing grid performance and maximizing the use of variable renewable energy resources,” in Proc. 2009 EVS24 Int. Battery, Hybrid and Fuel Cell Electric VEHICLE Symposium

[10] K. Clement, E. Haesen and J. Driesen, “Coordinated charging of multiple plug-in hybrid electric vehicles in residential distribution grids,” in Proc. 2009 IEEE Power Systems Conf. and Exposition, pp. 1-7

[11] Z. Ma, D. Callaway and I. Hiskens, “Decentralized charging control for large populations of plug-in electric vehicles,” in Proc. 2010 IEEE Conf. on Decision and Control, pp. 206-212

[12] A. Aabrandt, P.B. Andersen, A.B. Pedersen and S. You, “Prediction and optimization methods for electric vehicle charging schemes in the EDISON project,” to be presented at 2012 IEEE PES Conference on Innovative Smart Grid Technologies

[13] B. Lunz, H. Walz and D. U. Sauer, “Optimizing vehicle-to-grid charging strategies using genetic algorithms under the consideration of battery aging,” in Proc. 2011 IEEE Conf. on Vehicle Power and Propulsion Conference, pp. 1-7

[14] F. Marra, M. M. Jensen, “Power quality issues into a Danish LV grid with electric vehicles,” presented at IEEE EPQU, Lisbon, Portugal, 2011

[15] C. Søndergren, N. C. Bang, C. Hay, M. Togeby, and J. Østergaard, "Electric vehicles in future market models, Edison project," Danish Energy Association et al, DK, Tech. Rep. 2011.

[16] L. Christensen, “Edison project report: WP1.3 Electric vehicles and the customers,” DTU Transport, DK, Tech. Rep. 2011.

[17] www.axeonpower.com [18] S. You and C. N. Rasmussen, "Generic modeling framework for

economic analysis of battery systems," presented at the IET Renewable Power Generation Conf., Edinburgh, UK, 2011.


A.5 Coordination strategies for distribution gridcongestion management in a multi-actor, multi-objective setting

This paper was presented at IEEE Innovative Smart Grid Technologies (ISGT)conference, Berlin, Oct 2012.

1

Coordination Strategies for Distribution GridCongestion Management in a Multi-Actor,

Multi-Objective SettingPeter Bach Andersen, Junjie Hu, Kai Heussen

Abstract—It is well understood that the electric vehicle asa distributed energy resource can provide valuable servicesto the power system. Such services, however, would have toco-exist with hard constraints imposed by EV user demandsand distribution grid operation constraints. This paper aimsto address the interactions between the stakeholders involved,mainly considering the distribution grid congestion problem, andconceptualize several approaches by which their diverse, poten-tially conflicting, objectives can be coordinated. A key aspect to beconsidered is the relationship between the operational planningand the handling of real-time events for reliable grid operation.This paper presents an analysis of key stakeholders in terms oftheir objectives and key operations. Three potential strategiesfor congestion management are presented and evaluated basedon their complexity of implementation, the value and benefitsthey can offer as well as possible drawbacks and risks.

Index Terms—Electric vehicle integration, Distribution grid,Congestion management, Smart charging

I. INTRODUCTION

Grid integration of electric vehicles, distributed generation,and other distributed resources has been a driver for a rangeof smart grid research activities. Here, the field of intelligentelectric vehicle (EV) integration is aimed at minimizing theadverse effects of introducing electric vehicles into the powersystem and maximizing the value for EV owners, the powersystem, and society as a whole.

A large part of intelligent EV integration research has beenaimed at such topics as optimal charging of electric vehiclesin term of charging cost [1]–[3], enabling renewable energy[4]–[6] as well as providing ancillary service to the powersystem [1], [7]–[9]. Such studies have primarily been aimedat system-wide power services and energy markets while notconsidering the distribution network. Concurrently, studieshave been carried out that look at charging management solelyfor the purpose of avoiding distribution level grid congestion[10]–[14].

Lately, research done in [15], [16] have been striving tocoordinate these objectives, i.e., to optimize the utilizationof electric vehicles while still respecting the hard constraintsimposed by consumers’ needs and distribution operation con-straints. In [15], a conceptual framework consisting of boththe technical grid operation and a market environment wasproposed to integrate EVs, the activities of all the actors

This work was supported in part by the iPower project.Peter, Junjie and Kai are with the Center for Electric Technology, Technical

University of Denmark, Denmark. Email: {pba,junhu,kh}@elektro.dtu.dk

including fleet operator (FO), distribution system operator(DSO) and consumers are described and the simulation resultsindicate that smart charging can maximize the EV penetrationwithout exceeding grid constraints. However, further researchon the coordination between FO and DSO and the interactionbetween FOs and consumers are not addressed clearly. Afurther development can be seen in [16], in which a complexscheduling problem involving consumer, fleet operatoor andDSO were analyzed. The results shows that both power andvoltage constraints due to electric vehicle charging can beavoided while the FO and consumer can achieve the objec-tive of minimizing charging costs and fulfilling the drivingrequirements. This approach requires a somewhat complexcoordination between DSO and FO but can potentially delivera very good solution in terms of optimal grid utilization andsafety.

This paper aims to add to the existing research by addressingthe interactions between the various actors and conceptualizeseveral approaches, by which their diverse, potentially con-flicting, objectives can be coordinated with respect to theoperational constraints of the low voltage distribution grid.A key aspect to be considered is the relationship between theoperational planning done by the actors and the handling ofreal time events which is vital for the DSO and the distributiongrid that it represents.

While this paper focuses specifically on the case of EVintegration, the coordination strategies presented, aiming atcongestion management in general, can to a large extendbe translated to a more generic demand side managementperspective.

The remainder of this paper is organized as follows: Sec-tions II presents three key stakeholders along with theirobjectives and operational tasks. In Section III a full mapof the operations identified is presented and Section IV thenexpends the map in the examination of three different coor-dination strategies. Finally, key contributions are summarizedand discussed in Section V.

II. ACTORS: OBJECTIVES AND OPERATIONAL TASKS

An overview of the actors, the grid and the main controloperations is presented in Figure 1. The figure conveys howthe actors’ operations are coupled through interactions via a)a common physical infrastructure, b) control relations andc) other information exchange. The coordination of theseoperations needs to reflect each actor’s objectives as well

2

Fig. 1. Actors (stakeholders), problem domain and main information andcontrol flows.

as operational constraints. In this problem formulation, wefocus on describing the following key stakeholders and theirobjectives.

• Distribution System Operator (DSO),• Fleet Operators (FO),• Customers (controllable loads / EVs).Other relevant influencers include the transmission system

operator (TSO), other market actors and conventional demand.Their influence is conveyed via control signals, market prices,and physical network utilization, respectively. They do nothave to be considered here explicitly as their role with respectto the distribution level is encapsulated via the DSO and FO.

In the following, these key actors are described in termsof objectives and the operations performed to satisfy theseobjectives.

A. Distribution System Operator

The main purpose of the distribution grid is to enablereliable power delivery to customers at a low-voltage level.Grid operation by the DSO is therefore aimed at effectivelybalancing two main objectives a) reliable grid operation and b)low cost of operation. We identify the following value driversfor a DSO:

1) Grid component investments,2) Capacity utilization factor,3) Component lifetime,4) Operation cost (incl. resistive losses),5) Instrumentation and automation efforts.Provisioning of distribution grid transfer capacity is planned

to be sufficient in all cases, that is, capacity is provided by thestandard of annual peaks, plus safety factors for anticipateddemand increases. In practice this means that distributiongrids tend to have a relatively low utilization factor. On theother hand, the distribution grid planners calculate with a high

‘diversity factor’: it could be safely expected that due to theindependent nature of most electricity consumption would leadto a smoothing effect that would reduce the absolute peak. Asa result, secondary transformers in the distribution grids canbe expected to be dimensioned at lower capacity than the totalcurrent capacity of all connected households.

The operating state of the distribution grid is limited by thefollowing operation constraints:

• Voltage limits (voltage quality),• Thermal limits of cables & transformers,• MVar bands (interface to TSO), or• Protection settings.

In this paper the focus is on the distribution grid’s ability totransfer active power.

1) Congestion management: The term ‘congestion’ in dis-tribution grids refers to a situation in which the demand foractive power transfer exceeds the transfer capability of thegrid. As the electricity grid cannot physically get congested,the term subsumes the complex mapping of the above men-tioned grid constraints to the network active power transfercapacity as seen for each connection point and the need fordeferring demand (or generation1). Whereas the constraintslisted above are specified in terms of limits for specificparameters (voltage, current, reactive power, active power),they all may influence the active power transfer capabilityavailable at a connection point. Their mapping is non-trivial,as it depends on properties of the physical infrastructure,characteristics of consumption devices and built-in controlbehaviours required by the respective grid code.

In general, the term ‘congestion mitigation’ can then beassociated with two types of strategies: a) to (locally) increasethe transfer capacity by means of reactive power and voltagecontrol and b) by coordinating the throughput via deferral orcurtailment of demand [17]. Both strategies aim at increasingthe utilization factor of the distribution grid.

Here, the term ‘congestion management’ explicitly refers tostrategies of type b), which aim at the coordination of activepower demand with respect to congested grid locations. Itcan be assumed that available strategies of type a) will beexhausted before type b) strategies are applied. Building onthe proposal in [17], the base case for congestion mitigationwill be considered active power curtailment.

2) Distribution System Operation Today: DSO tasks inconventional system operation, are mostly focused on ‘off-line’ tasks related to asset management and maintenance.Distribution systems today tend to be weakly monitored ascompared to transmission grids, and controlled in a decentral-ized fashion on the basis of preconfigured local controls (e.g.by means of grid codes and protection settings). Supervisorycontrol is then reduced time-of-day controlled adaptation ofcontrol settings, configuration management in response tooutages and maintenance related challenges.

Key Operations:• Grid dimensioning (incl. contigency planning and load

curve estimation),

1For the remainder of this paper, the perspective of distributed generationis implied.

3

• Maintenance and outage related topology reconfiguration,• Adjustment of transformer taps,• Fuses and relay operation,• Fault-analysis and repair.3) Operations in active distribution grids: To illustrate a

future operation scenario with a higher level of automation,it is considered how the above operations can be extendedwith additional online- and data intensive operations. In orderto identify and solve congestion problems, the DSO requiresadditional measurement equipment and/or technology enablingthe anticipation of load patterns and grid ‘bottlenecks’.

Key Operations for DSO congestion management in ‘ad-vanced’ distribution grids:

• Demand forecasting• Grid state estimation• Online grid measurements• Real-time intervention in case of unexpected deviations

challenging grid reliability• Meter data collection and aggregation

economically and reliably and shows a relation between VPPsand DSO. In this control system, several families are suppliedunder one feeder and they own controllable devices, i.e,electric vehicles, besides some conventional load, such as light,TV etc. For these controllable devices, they are divided intotwo groups according to the method controlled by the VPP,one group is directly controlled by the VPP, which meansan extra cards or relays are installed on the user’s device,and the VPP can turn on/off the devices; another group iscontrolled by price, in which the devices are assumed to beprice-responsive. VPP starts to make an energy schedule forits customers with the purpose of minimizing the electricitycost and meanwhile fulfilling their requirement. This problemcan be formulated as a linear programming or dynamicalprogramming way [1], [2]. The congestion problem mayfirst happened during the scheduling making, this problemshould be solved by the coordination between DSO and VPPs.After the charging schedule was set up, ideally, the users areexpected to totally follow the schedule. However, in general,deviation may happen. With the purpose of avoiding thepossible congestion (happened again in real time), DSO willmonitor the system’s operation conditions dynamically andcoordinate with VPPs. The following subsection will discussthe mechanism of solving these congestion problems.

B. Fleet Operators

The fleet operator (FO) is a commercial entity that aggre-gates a group of EVs in order to actively integrate them intothe power market, and in so doing, utilizing their chargingflexibility to meet a financial goal. The financial goal couldbe to achieve savings on the purchase of energy or makeearnings by selling ancillary service products or, possibly, acombination of the two.

A FO follows the concept of a ’virtual Power Plant’which was first introduced to allow market participation fordistributed energy resources.

In the current European power and energy markets, the FOcould be a retailer with either a load balance or production

balance responsibility, depending on the market/service thatthe FO would address.

The value drivers for a FO are:1) Maximize profits or minimize costs by participating in

markets.2) Providing services (cost reductions, convenience etc.) that

will attract EV owners as customers.Due to the participation in markets and customer services,

the FO is subject to operating constraints defined by contrac-tual commitments:

• Market schedule (energy/h)• Customer demand (driving needs)• TSO driven ancillary service requirements (e.g. reserve

capacity)How the economic value obtained through the market is

shared with the customers would be business case specificto the FO. It is also assumed that the FO would maintain aService Level Agreement (SLA) with its customers that woulddictate the degree to which it may control and manipulate theEV charging patterns to achieve goals other than customerdriving. This would represent a trade-off between energysavings and EV driving availability that should be understoodand accepted by the customer.

The operations of the fleet operator can be divided into fleetlevel operations and individual level operations as follows.Fleet level:

• Selection of market products and services• Contracting• Market/service forecasting

Individual level:• Customer SLA management• Driving pattern prediction

C. Customers

The customers, here EV owners and drivers, are not as-sumed to be particularly interested in grid issues. Their mainvalue drivers are expected to be:

1) Availability of EV for driving2) Total cost of ownership/energyIt is assumed here, that the customer will opt for conve-

nience and delegate most of the charging control to the FO.The customer is expected to rely on the frame conditionsexpressed in the SLA for the daily charging managementfor ’typical’ and predictable driving patterns. An optionalfeature would be to let the customer communicate his or hersexact driving intentions to the FO. This would strengthen theFO ability to utilize the specific EV’s flexibility. The mainoperations of the customer, besides transportation, would thenbe:

• Accept, and possibly modify, the SLA with the FO.• Inform the FO of any non-typical driving needs.

III. MAP OF OPERATIONS

The operations outlined above will in this section be mappedgraphically to enable an analysis of different coordinationstrategies.

4

Fig. 2. Map of base-case operations.

A. Analysis FrameworkIn this section a classification approach is introduced, based

on the understanding that the coordination approach is both anautomation problem and a market design problem.

A widely accepted hierarchical decomposition of processcontrol into four functional levels describes integrated indus-trial automation [18]:

• Level 4: plant(s) management• Level 3: production scheduling and control• Level 2: plant supervisory control• Level 1: direct process control

This level-hierarchy is associated with several characterizingparameters, including e.g. time scale, time resolution, planninghorizon, and hierarchical dependency of objectives. No singleone of these parameters can be considered directly decisivefor forming the levels, but together they generate the needfor distinguishing qualitatively different levels of automation.The hierarchical dependency of objectives, i.e. that one level ishigher and another is lower in ordering, is associated with themeans-ends structure of objectives: A higher-level objective isbroader in scope and more closely associated with the businessobjectives of the respective process, and thus ‘higher’ in thevalue chain; a lower-level objective, in contrast, is there tosupport and enable other process functions.

In the present multi-actor, multi-objective setting, the sin-gle hierarchy does not hold: different actors have differentobjectives, and yet they must interact with respect to the sameprocess. The present means-ends perspective shall be strippedfrom the automation hierarchy, to allow for a high-level mapof operations. Key elements to be captured in the new mapare:

• Key operations and their allocation to actors• The association of operations with a time scale including

a distinction of operational and administrative functions

• The possibility to map interaction sequences betweenoperations

Removing the means-ends perspective, we are left with amostly time-driven decomposition. We consider the followingfundamental stages:

I. Offline PlanningII. Online Scheduling

III. Real-time Operation (Execution)IV. Offline Settlement

These stages model a logical sequence: each stage is basedon a completion of the previous stage. The timing aspect is notessential here as certain types of operation can be performedfaster with improved technology. The stages are characterizedin the following:

Settlement is about the aftermath: recordings (measure-ments, sent commands, etc.) of executed operations are con-solidated and (financial) responsibility is allocated. The op-eration stage is about pure execution in real-time. Plans areonly executed, and unplanned events occur and physical aswell as automatic controls respond without deliberation. The’online’ scheduling stage can in time be closely coupled withoperation (e.g. reactive scheduling with a 5min resolution) orextend hours or days ahead of it. Scheduling is the stage inwhich available resources are best known and the platformfor execution is to be prepared. Finally, the first stage, herecalled ’planning’ has been distinguished from scheduling inthe same fashion as unit commitment is distinguished fromdispatch: Depending on the specific coordination strategy, wedistinguish operations that can be coordinated in a ad-hocfashion and those that provide the basis for such ad-hoc deci-sions. Due to these clear distinctions, the framework supportsthe discussion of interactions between key operation tasks forcross-stakeholder coordination for the complete process. Asthe operations can be associated with operation objectives of

5

the respective stakeholder, this map allows for an analysis ofthe incorporation of the respective value drivers by a givencoordination strategy. This ’horizontal’ level means that theoperations have to be considered at the same level of abstrac-tion. A ’vertical’ perspective would unfold more details of theoperations, eventually also revealing physical interactions [19].The goal is to analyze the benefits and trade-offs involved inspecific coordination strategies. Value is hereby understood ina generic sense as to contribution to a stakeholder’s objectives.Given the Operation-Stakeholder allocation and the analysisof value drivers, a similar framework can be employed to alsoanalyse value-network constellations, as exemplified in [20].However, that type of analysis is beyond the scope of thispaper.

B. Base Case Map

To establish a firm foundation for the analysis of differentmarket-based coordination strategies, we identify a base casewith a minimum set of operations that will be common toall considered congestion-coordination strategies (Figure 2).This base-case maps out the operations DSO, FO and EVowner would be required to execute in either of the coordinatedcongestion management schemes.

The base case uses the following assumptions:- As discussed in [15] and [19], the introduction of con-

trollable demand with significant power capacity such as thatof electric vehicles implies a significant risk for distributionassets. To avoid potentially harmful charging configurations,we include the concept of an ’emergency brake’ in all EVcharging post: it enables the unconditional interruption of EVcharging on request by the DSO. It could be implemented onthe basis of a ’keep-alive’ signal, the failure of which wouldimmediately interrupt the EV charging process

- The maps allocate all optimization and coordination intel-ligence to the FO. It is understood that many of the operationscould be implemented using distributed algorithms e.g. in theelectric car or charging post.

- Vehicle-to-Grid (V2G) is not considered in this publica-tion. The technology’s potential for congestion relief and itsimpact on power quality are, however, relevant for conges-tion management and should be further addressed in futurepublications.

IV. COORDINATION STRATEGIES FOR CONGESTIONMANAGEMENT

Approaches to the congestion problem are outlined and thenclassified and analysed using the map described in the previoussection. All three strategies represent very new approaches todistribution grid congestion management and none of themhave been investigated in very great detail.

The strategies investigated are:• Distribution grid capacity market• Advance capacity allocation• Dynamic grid tariffFor each approach a new map is drawn where operations

required specifically for the strategy in question are presented

Fig. 3. Distribution Grid Capacity Market.

in bold. Shared supporting operations beyond the main traceof operations have been omitted for compactness.

To describe how technically and administratively demandingit would be for a DSO or FO to implement and operate the newprocedures required by the coordination strategy the parametercomplexity is used. A second parameter value denotes the de-gree to which the strategy would help the stakeholder achieveits operational goal. Finally, the parameter risk describespotential problems associated with the respective strategy incontext of a currently uncertain external environment.

A. Distribution grid capacity market

As proposed in [21], this strategy would require a newmarket for trading distribution grid capacity. For this paper theterm ‘Distribution Grid Capacity Market’ is used; Also a newrole ’market operator’ is introduced which is responsible formarket operation. The FO will submit requests for their ’aggre-gated schedule’ consisting of their scheduled consumption foreach node (aggregated capacity), in response they will receivea price for each node, reflecting the respective congestion, andare requested to update their charging schedules. The processis iterated until all constraints are satisfies. The concept usedin this strategy can be found in a similar form for the powertransmission system [22].

1) Operation sequence:• First, the FO will make an aggregated energy schedule

for EV owners based on its objectives. Afterwards, thisaggregated schedule will be sent to the market operator.

• The Market operator will generate a price for the grid ca-pacity according to the schedules. This price is associatedwith the power difference between the sum of scheduledpower and upper power limits of the grid.

• FOs would inform the market operator of their newenergy schedule under the initial price. The schedule canbe calculated based on the marginal value of a utilityfunction, e.g., cost function in term of the power deviationor satisfaction degree with the ’preference difference’(the difference between energy schedule after congestionmanagement and energy schedule before congestion man-agement).

• The market operator then determines whether the distri-bution grid is overloaded or underutilized and comes upwith a new corresponding price. After a certain number

6

Fig. 4. Advance Capacity Allocation.

of iterations, the price will eventually converge andaccepted by all FOs, which establishes a binding chargingschedule.

2) Evaluation: Complexity:With this strategy, a new market is required which means

that the corresponding platform for trading grid capacity needsto be designed and implemented. Also, new communicationflows are needed to support the market operations. The marketitself will be rather complex to establish and operate.

A lot of complexity is transfered from DSO to the capacitymarket. Here the DSO will be required to provide the measuredand estimated power information to the market operator. TheFOs will take on the task of trading capacity and reschedulingthe energy consumptions for their customers etc..

Value: With this new market, FOs will have more flexibilityto trade and utilize the grid capacity of a distribution system.If the market and capacity information is reliable and well-designd it will ease the operation of the DSO, enable acomparatively high utilization factor and reliable schedules forFOs as well. A further benefit is that no actual consumptioninformation is revealed to other market parties, as only acommon congestion price is established per node.

Risk: It must be guaranteed that all FOs adhere to the rulesof the market. Another risk lies in the algorithms used toarrive at prices based on utility functions i.e. the computationalrequirements and time needed for a solution to converge.

B. Advance Capacity Allocation

The simple concept behind this strategy is that the DSOcould identify and pre-allocate available capacity by defininga conservative static capacity limit (kW) for each feeder-linebased on the capacity rating of the respective transformers andcables and the expected conventional load curves. The EV-equipped households attached to a certain feeder-line wouldthen be given a certain share of available capacity which wouldbe allocated to the FO representing them. To avoid inefficientutilization of available grid capacity due to unused capacityshares, a second step is added to the strategy where FOs cantrade their allocated capacity in an over-the-counter manner.

1) Operation sequence:

• The ’Contracting for capacity sharing’ operation wouldinvolve letting the DSO know the mapping between grid

connected EV-equipped customers and FOs and thendetermining how capacity is shared.

• During the scheduling stage, the DSO would via grid loadforecasting estimate the available capacity and communi-cate this to the FOs as defined in the contracts establishedin the planning stage. After having received its share, theFO could then optionally engage in capacity trading withother FOs operating on the same feeder.

• The DSO should be informed of the bilateral capacitytrading so that, in case of violations (i.e. total load ob-served from EV charging in specific part of grid exceedssum of allocated shares) penalties for violations can beappropriately placed at the responsible FO.

• Finally the strategy would involve settlement both be-tween DSO and the individual FO and possibly aninternal settlement between the FOs that engaged in thebilateral trade.

2) Evaluation: Complexity: Here, rather than dimension-ing the physical characteristics of the grid depending on loadprofiles and simultaneity factors, the DSO would limit thecontrollable load based on the physical characteristics of thegrid. In addition to the location-based grid capacity the DSOwould also need to map each grid customers endpoint to anassociated FO. There is also some complexity in how theFOs will trade capacity internally and how violations of gridcapacity will be dealt with in the settlement stage when tradinghas been involved.

Value: This strategy represents a rather simple coordinationmechanism between FOs and DSOs. The DSO is only requiredto communicate a single value (capacity) to each FO and isthen removed from the equation until the settlement stage. Thiswill simplify the responsibilities of the DSO considerably andleave the detailed capacity allocation to the entities directly incontrol of EV charging i.e. the FOs. There are also advantagesto the FO since it will see a guaranteed capacity, free fromstochasticity, early in the planning stage. Early information isvaluable to an FO attempting to optimize charging to meet avariety of goals such as market services and individual drivingneeds.

Risk: There is the risk that a single kW limit set-point pergrid node is too crude a mechanism to handle thermal loading- any unexpected change in base load during operation mayvoid the DSO’s estimation of capacity shares which has beenhanded out to the FOs during scheduling. The risk in thisapproach also lies in the effectiveness and reliability of theFO bilateral capacity trading. If the FOs can not be trustedto handle the management and trading of capacity amongthemselves, there will be the need of a more formal framework,e.g a market, and new definitions of responsibilities, such asthe balance responsible parties seen in the energy market.

C. Dynamic Grid Tariff

In this solution, the distribution system operator generatesa time and grid-location dependent price for grid usage basedon expected nodal consumption levels.

The DSO anticipates the size and the price-responsivenessof the load at critical grid nodes and calculates the price to

7

Fig. 5. Dynamic Grid Tariff.

optimally reflect the expected congestion problem. FOs willthen see a dynamic nodal tariff and can make an optimalschedule with respect to the e.g. spot price and dynamic gridtariff.

The method considered here has been presented in [23].1) Operation sequence: The key operation aspects for this

coordination strategy are outlined in Figure 5.• In the planning stage, a distribution system operator

would create models for the price-sensitivity of relevantdemand clusters. These models would be updated ona regular basis based on learning from smart meterfeedback.

• In the scheduling stage the forecasted demand, gridsituation and present spot market prices will be employedto calculate appropriate branch prices for distribution gridutilization.

• The dynamic tariff is published to subscribers. Theadapted branch prices are received by the fleet operatorand employed to compute an optimal charging plan.

• During the operation stage, the charging schedule is ex-ecuted. In case of severe underestimation or fluctuationsof the actual demand, DSO controlled interruptions mayoccur in real-time.

• For settlement the timed consumption data is collected bythe responsible DSO and the published prices will thenbe employed to bill the actual grid usage individually.

2) Evaluation: Complexity: The main characteristic fea-ture of this approach is the simplicity of the interactions andalso the simplicity of integrating simple prices in distributiongrids.

The implementation complexity is high on the side ofthe DSO. This scheme cannot be established safely withoutinterruptability of the vehicle charging.

For the Fleet Operator–Consumer interaction, the establish-ment of a satisfactory service quality may require a specialattention to potential bottlenecks in the system from the sideof the Fleet Operator.

Value: As compared to the base case, this model enablesan increase of the grid utilization factor. The small number ofparticipants at a feeder level means that random behaviour(fluctuation consumption level) might be stronger than theprice sensitivity of the controllable demand. Even though theincrease of the utilization factor is therefore highly uncertain,the simplicity of the approach could justify its implementation.

TABLE ISTRATEGY OVERVIEW

Actor Complexity Value RiskDistribution Grid Capacity Market

FO High High LowDSO High High Low

Advance Capacity AllocationFO Medium High LowDSO Low Medium Medium

Dynamic Grid TariffFO Low Low HighDSO Medium Medium High

For fleet operators and consumers, the benefits are alsoindirectly associated with the increased grid utilization. Afurther benefit can be seen in the flexibility this approach offerswith respect to integrating other flexible demand units, as theprice, in theory, could interpreted by any unit.

Risk: It is unclear whether a meaningful price-sensitivityof demand can be established.

There is a risk that there is no ‘right’ price to avoidoverloading, if a sufficient number of EVs is connected tothe same feeder, there is no way for them to negotiatecapacity utilization in the given framework. Due to the re-quired interruptability, the high chance for unplanned charginginterruptions also implies an additional risk is on the side ofthe Fleet Operator / EV Owner.

V. DISCUSSION AND CONCLUSION

This paper has investigated the concept of congestionmanagement for distribution grids, detailing the operationsand interactions of two main stakeholders in three differentcoordination strategies. The purpose of the analysis was tohighlight the cross-actor dependencies that each such strategyentails along the operation timeline, and thus to globally assesscomplexity, value and risk for each strategy.

Table I summarizes these evaluation parameters across thestrategies and stakeholders. In this table the customer isrepresented by the FO.

The ‘distribution grid capacity market’ is expected to offerhigh value and low risk for both FO and DSO assuming aformalized, optimal and secure framework supplied by a well-designed market. such a market, however, may represent themost complex strategy to implement, which may hinder ordelay its real-life implementation.

‘Advance grid capacity’ is relatively easy to implement,but would require over-the-counter trading to efficiently useavailable grid capacity. The strategy removes complexity fromthe DSO but some risk may have to be managed due to thebilateral FO trading and the advance capacity allocation mightrequire more conservative estimates of the available capacity.The FO gains high value from early information on capacityavailability.

‘Dynamic tariffs’ would also be easier to implement thana capacity market, but may prove challenging to the fleetoperator due to added uncertainty and a possible conflict withsystem-wide smart charging schemes. It could also imposesome extra risk for the DSO to rely on prices rather than hardcapacity limits when considering individual feeder lines.

8

A few general observations can be summarized as:• Grid considerations will have to co-exist with other

objectives in the charging management of EVs.• Coordination and information exchange in earlier stages

can reduce complexity and benefit both the FO and DSO.• There may be a trade-off between ease of strategy

implementation and optimality. A compromise may benecessary for the first real-life implementations.

• A suitable strategy for coordination between DSOs andFOs will improve each stakeholders ability to reach itsobjectives considerably.

The analysis framework developed in this paper can beconsidered sufficiently generic for analysing operations withrespect to other distributed resources.

An important mechanism included in this paper is the ‘real-time intervention’ functionality used by the DSO. This last-resort ability to directly and immediately reduce or disconnectcharging may be a prerequisite for the deployment of effectivecoordination strategies.

In the end, it is hoped that this paper contributes to a betterunderstanding of the multifaceted challenge of EV chargingand helps the development of open, robust, and meaningfulstrategies for low voltage grid congestion management.

VI. REFERENCES

[1] N. Rotering and M. Ilic, “Optimal charge control of plug-in hybridelectric vehicles in deregulated electricity markets,” IEEE Transactionson Power Sy, 2010.

[2] O. Sundstrom and C. Binding, “Optimization methods to plan thecharging of electric vehicle fleets,” in Proceedings of the InternationalConference on Control, Communication and Power Engineering, 2010,pp. 28–29.

[3] T. Kristoffersen, K. Capion, and P. Meibom, “Optimal charging ofelectric drive vehicles in a market environment,” Applied Energy, 2010.

[4] F. Tuffner and M. Kintner-Meyer, “Using electric vehicles to meetbalancing requirements associated with wind power,” Pacific NorthwestNational Laboratory (PNNL), Richland, WA (US), Tech. Rep., 2011.

[5] J. Pecas Lopes, P. Rocha Almeida, and F. Soares, “Using vehicle-to-gridto maximize the integration of intermittent renewable energy resourcesin islanded electric grids,” in Clean Electrical Power, 2009 InternationalConference on. IEEE, 2009, pp. 290–295.

[6] J. Lopes, F. Soares, P. Almeida, and M. da Silva, “Smart charging strate-gies for electric vehicles: Enhancing grid performance and maximizingthe use of variable renewable energy resources,” in EVS24-The 24thInternational Battery, Hybrid and Fuel Cell Electric Vehicle Symposium& Exhibition, Stavanger, Norway, 2009.

[7] W. Kempton and J. Tomic, “Vehicle-to-grid power implementation: Fromstabilizing the grid to supporting large-scale renewable energy,” Journalof Power Sources, vol. 144, no. 1, pp. 280–294, 2005.

[8] ——, “Vehicle-to-grid power fundamentals: Calculating capacity and netrevenue,” Journal of Power Sources, vol. 144, no. 1, pp. 268–279, 2005.

[9] S. Han, S. Han, and K. Sezaki, “Development of an optimal vehicle-to-grid aggregator for frequency regulation,” IEEE Transactions on SmartGrid, vol. 1, no. 1, pp. 65–72, 2010.

[10] G. Heydt, “The impact of electric vehicle deployment on load manage-ment straregies,” Power Apparatus and Systems, IEEE Transactions on,no. 5, pp. 1253–1259, 1983.

[11] ——, “The impact of electric vehicle deployment on load managementstraregies,” IEEE Transactions on Power Apparatus and Systems, no. 5,pp. 1253–1259, 2007.

[12] R. Green II, L. Wang, and M. Alam, “The impact of plug-in hybridelectric vehicles on distribution networks: A review and outlook,”Renewable and Sustainable Energy Reviews, vol. 15, no. 1, pp. 544–553, 2011.

[13] J. Lopes, F. Soares, and P. Almeida, “Identifying management pro-cedures to deal with connection of electric vehicles in the grid,” inPowerTech, 2009 IEEE Bucharest. IEEE, 2009, pp. 1–8.

[14] K. Clement-Nyns, E. Haesen, and J. Driesen, “The impact of chargingplug-in hybrid electric vehicles on a residential distribution grid,” PowerSystems, IEEE Transactions on, vol. 25, no. 1, pp. 371–380, 2010.

[15] J. Lopes, F. Soares, and P. Almeida, “Integration of electric vehicles inthe electric power system,” Proceedings of the IEEE, vol. 99, no. 1, pp.168–183, 2011.

[16] O. Sundstrom and C. Binding, “Flexible charging optimization forelectric vehicles considering distribution grid constraints,” Smart Grid,IEEE Transactions on, vol. 3, no. 1, pp. 26–37, 2012.

[17] T. Sansawatt, L. Ochoa, and G. Harrison, “Integrating distributed gen-eration using decentralised voltage regulation,” in Power and EnergySociety General Meeting, 2010 IEEE, july 2010, pp. 1 –6.

[18] D. Popovic and V. Bhatkar, Distributed computer control for industrialautomation. CRC, 1990.

[19] K. Heussen, “Control architecture modeling for future power systems,”Ph.D. dissertation, Technical University of Denmark, 2011.

[20] K. Kok, Z. Derzsi, J. Gordijn, M. Hommelberg, C. Warmer, R. Kam-phuis, and H. Akkermans, “Agent-based electricity balancing withdistributed energy resources: A multiperspective case study,” in Pro-ceedings of the 41st Annual Hawaii International Conference on SystemSciences, 2008.

[21] B. Biegel, P. Andersen, J. Stoustrup, and J. Bendtsen, “Congestionmanagement in a smart grid via shadow prices,” in Proceedings of the 8thIFAC Symposium on Power Plant and Power System Control, Toulouse,France, Sep. 2012.

[22] S. Stoft, Power system economics. IEEE press Piscataway, NJ, 2002.[23] N. O’ Connel, Q. Wu, J. stergaard, A. Nielsen, S. Cha, and Y. Ding,

“Electric vehicle (ev) charging management with dynamic distributionsystem tariff,” in proceeding of 2011 ISGT, 2011.

VII. BIOGRAPHIES

Peter Bach Andersen has a M.Sc. in Informaticsand is currently finalizing a PhD at the Departmentof Electrical Engineering within the Technical Uni-versity of Denmark (DTU). His research focus is onsmart grids in general but with a special emphasis onelectric vehicle integration. Peter is an IEEE studentmember.

Junjie Hu received the Master of Engineeringin Control Theory and Control Engineering fromTongJi University, ShangHai, China, in 2010. Cur-rently, he is a PhD student at the Department of Elec-trical Engineering within the Technical University ofDenmark (DTU).His research focuses on integratingcontrol policies on controllable load, mainly electricvehicle, for active power distribution system, ofwhich the control policies are direct control andindirect (price) control. Junjie is an IEEE studentmember.

Kai Heussen is assistant professor at the TechnicalUniversity of Denmark, where he also obtained hisPhD on Control Architecture Modeling for FuturePower Systems. He received his Dipl.Ing. in En-gineering Cybernetics in 2007 from University ofStuttgart. His current research focus is the design ofheterarchical and service-oriented control architec-tures for distributed control of power systems, withspecial attention to functional modeling and decisionsupport for automation design.

Coordinated charging of electric vehicles for congestion prevention in thedistribution grid 135

A.6 Coordinated charging of electric vehicles forcongestion prevention in the distribution grid

This paper is published in IEEE transactions on smart grids, volume 5, Issue 2,page 703-711, March 2014.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON SMART GRID 1

Coordinated Charging of Electric Vehicles forCongestion Prevention in the Distribution Grid

Junjie Hu, Student Member, IEEE, Shi You, Member, IEEE, Morten Lind, andJacob Østergaard, Senior Member, IEEE

Abstract—Distributed energy resources (DERs), like electric ve-hicles (EVs), can offer valuable services to power systems, suchas enabling renewable energy to the electricity producer and pro-viding ancillary services to the system operator. However, thesenew DERs may challenge the distribution grid due to insufficientcapacity in peak hours. This paper aims to coordinate the valu-able services and operation constraints of three actors: the EVowner, the Fleet operator (FO) and the Distribution system op-erator (DSO), considering the individual EV owner’s driving re-quirement, the charging cost of EV and thermal limits of cables andtransformers in the proposed market framework. Firstly, a the-oretical market framework is described. Within this framework,FOs who represent their customer’s (EV owners) interests will cen-trally guarantee the EV owners’ driving requirements and pro-cure the energy for their vehicles with lower cost. The congestionproblem will be solved by a coordination between DSO and FOsthrough a distribution grid capacity market scheme. Then, amath-ematical formulation of the market scheme is presented. Further,some case studies are shown to illustrate the effectiveness of theproposed solutions.

Index Terms—Congestion management, distribution grid, elec-tric vehicles, optimal charging schedule.

I. INTRODUCTION

A. Aggregated Charging of EVs

O THER THAN fulfilling its basic transport function, EVs,as smart grid assets, can provide a large number of valu-

able services, e.g., meeting the balancing requirements for en-ergy suppliers with stochastic renewables, providing regulationservices to the system operators, and modifying the demandcurves to defer the network expansion, etc., [1]–[7]. As a re-sult, a new business entity, namely EV fleet operator (FO) hasemerged which aims to capture the business opportunities byproviding the multiple services of EVs. Alternative names foran EV FO are used such as EV virtual power plant, EV aggre-gator or charging service provider. The new entities could be in-dependent or integrated in an existing business function of theenergy supplier; however they all share a list of commonalitiesas following:1) Same mission:• Guarantee driving needs of the EV owners;

Manuscript received January 10, 2013; revised June 13, 2013; accepted Au-gust 13, 2013. This work is supported by Danish iPower project (http://www.ipower-net.dk/). Paper no. TSG-00019-2013.The authors are with the Center for Electric Power and Energy, Depart-

ment of Electrical Engineering, Technical University of Denmark, 2800Kgs. Lyngby, Denmark. (e-mail: [email protected]; [email protected];[email protected]; [email protected]).Color versions of one or more of the figures in this paper are available online

at http://ieeexplore.ieee.org.Digital Object Identifier 10.1109/TSG.2013.2279007

• Coordinate and support the valuable services and oper-ation constraints of EV and power system operator;

• Maximize the renewable energy.2) Similar methods:• Implement centralized control/marketing method tomaximize business values [1]–[7];

• Optimize the charging process of EVs [6], [8]–[11].However, the function of the distribution grid may be chal-

lenged when FOs try to achieve these objectives, because theincreasing size and number of consumption units, e.g., EVs cancause problems in peak hours [12]–[16]. Mainly two issues areconsidered in the literature when discussing the possible chal-lenges in a distribution grid with increasing DERs, they arevoltage drops and thermal overloading of transformers and ca-bles. Focus in this study will be on the prevention of the thermaloverloading of the transformers and cables, which is also knownas congestion management. In the following of this section, wefirst give an overview on the centralized and market-based coor-dination strategies for gird congestion management. Then, themotivation and contribution of this study is presented.

B. Centralized Control of EVs for Grid CongestionManagement

Lately, research done in [17]–[19] have been aiming to co-ordinate these multiple objectives centrally, i.e., to optimize thecharging cost of EVs as well as respecting the hard constraintsimposed by EV owner needs and distribution grid operation. In[17], [18], a complex scheduling problem involving EV owners,FO and DSO is analyzed. The results show that both the FOand the EV owners can achieve the objectives of minimizingcharging costs and fulfilling driving requirements without vi-olating the grid constraints. This approach requires a complexinteraction between DSO and FO, but can potentially delivera very good solution in terms of optimal grid utilization andsafety. A conceptual framework consisting of both a technicalgrid operation strategy and a market environment is proposedin [19] to integrate EVs into the distribution systems, the activ-ities of all the actors including the EV owner, the FO and theDSO are described and the results indicate that smart chargingcan maximize the EV penetration without exceeding grid con-straints.

C. Market-Based Coordination Strategies of EVs for GridCongestion Management

Alternatively, several ways of solving the congestion problemhave been suggested from market perspective. Our previousstudy [20] has conceptualized several approaches to address the

1949-3053 © 2013 IEEE


2 IEEE TRANSACTIONS ON SMART GRID

distribution grid congestion according to their value and bene-fits which they can offer as well as possible drawbacks and risksand the complexity of implementation. A short description ofthe principles behind these strategies is given below:• Distribution grid capacity marketIn this method, FOs will submit power requests to DSOfor their aggregated energy/power schedule on each node(aggregated capacity); in response they will receive a pricefor each node which reflects the respective congestion, andare requested to update their energy schedules. The processwill terminate when all constraints are satisfied. The mech-anisms behind the market could be designed in many ways,such as uniform price auction mechanism [21], shadowprice based mechanism [22].

• Dynamical grid tariffIn this method [23], the DSO generates a time and grid-lo-cation dependent price for grid usage based on expectednodal consumption levels. The DSO anticipates the sizeand the price-responsiveness of the load at critical gridnodes and calculates the price to optimally reflect the ex-pected congestion problem. The FO will then get the dy-namic nodal tariff and make an optimal schedule with re-spect to the predicted spot price and dynamic grid tariff.

Besides, studies in [24], [25] investigate the coordinationmethods using a common price signal and the work [24], [25]share a lot of similarities. Mainly, both work use game theoryto formulate an EV/demand energy consumption schedulinggame. The actors are assumed to be cost-minimizing andcoupled via a common signal, i.e., a common electricity pricein [25] and total load of the distribution grid (a heavier gridloading means a higher price) [24]. The strategies are the dailyschedule of their consumption.

D. Motivation and Contribution of This Paper

Currently the dispatch is set only based on the day-aheadelectricity market and the end-users’ need for energy services.Traditionally, demand takes place when needed and the chal-lenges in the distribution grid created by the EV aggregationcould be solved by expanding the grid to fit the size and patternof demand. As an alternative, it is assumed that the distributiongrid company can benefit more by making the consumers shiftdemand consumption from one given period to another, afteridentifying the long-term marginal costs of reinforcement of thegrid. Considering the new opportunity, a new scheme using boththe day-ahead electricity market and the distribution grid stateto set the dispatch should be established, which can enable thepower system balance and the distribution grid congestion man-agement. In general, two schemes are qualitative analyzed andproposed [12]: 1) distribution grid congestion management first,then energy system balance or 2) vice versa. Both the merits anddemerits of these two strategies are well discussed in the report[12].The present study aims to investigate the coordination

strategy among DSO, FOs and EV owners based on the basicconcept in [20], [12], specifically, the conceptualized proposalof “distribution grid congestion management first, then energysystem balance” in [12] and the distribution grid capacitymarket in [20], by making them into concrete optimization

Fig. 1. The proposed distribution grid capacity market can be an integrated partof the current power markets.

problems and by showing detailed case studies. In this study, aframework is proposed which can minimize the charging costof EVs as well as respecting to the hard constraint imposed bythe EV owners and DSO. The proposed framework consists oflinear programming technology based optimal charging of EVsand shadow price mechanism based distribution grid capacitymarket.Fig. 1 illustrate the proposal of integrating the new proposed

distribution grid capacity market into the existing power mar-kets. It is emphasized that themarket scheme is flexible and scal-able, in the fig, the “distribution grid capacity market scheme”is placed in three position, which represent three different timeperiods, i.e., day-ahead and intra-day period for congestion pre-vention, real time period for congestion relief.There are mainly two research contributions in this study.

Firstly, we recommend and test a distribution grid capacitymarket set up enabling the distribution congestion prevention,in which multiple FOs are involved. Secondly, the proposedmarket framework is flexible and scalable in diverse controlschemes, such as the mechanism elaborated later specificallyfor the day-ahead period can be also adopted for the controlsituation in the period of intra-day and real time, the mecha-nism designed for EVs can be also used for other applianceswhich have the capability of altering their consumption patternwith limited impact on their primary energy service, such astheoretically controlled loads.The remainder of the paper is organized as follows: In

Section II, a general explanation is given on the methodologyfor congestion prevention, i.e., the proposed market framework.Section III mainly presents the mathematical formulation of thecongestion issues into a subgradient method based distributiongrid capacity market set up. Then several case studies areillustrated in Section IV to facilitate the understanding. Finally,discussion and conclusions are made in Section V.

II. METHODOLOGY FOR CONGESTION PREVENTION: SYSTEMARCHITECTURE, OPERATION PRINCIPLE

As discussed in [26]–[28], each distribution grid has a dif-ferent history, such as in some cases congestion is first expectedto emerge in the medium voltage grid, while in other grids thelow voltage grid is considered to be more critical. In this study,a low voltage grid is used for illustration purpose, but the pro-posed framework holds for medium voltage grid as well.


HU et al.: COORDINATED CHARGING OF ELECTRIC VEHICLES FOR CONGESTION PREVENTION IN THE DISTRIBUTION GRID 3

Fig. 2. A schematic of a low voltage active distribution system.

A. System Architecture, Framework Design

Fig. 2 presents a schematic of a low voltage distributionsystem, in which around 60–70 household consumers areconnected to a 10/0.4 KV transformer (a typical Danish case),mainly connected on one feeder. In this distribution system, it isassumed that the consumers own controllable devices, i.e., EVs,besides some conventional loads, such as light or TV. TheseEVs are divided into two groups as illustrated in Fig. 2. Onegroup is controlled by fleet operator-1 , another groupis controlled by fleet operator-2 . In this hierarchicaldistribution system, both FOs can schedule and control theircustomer’s electricity consumption directly. While on the FOlevel, the coordination between FOs and DSO is made throughthe distribution grid capacity market.Within the assumed system architecture, we propose a frame-

work consisting of four fundamental stages [20] to operate andcontrol this system. This framework is a fully charging pro-file management for integrating EVs into the distribution gridsmoothly.1) Energy schedule of the FOs without congestion manage-ment—Offline scheduling. Both and need topredict the energy requirements (driving patterns) of theircustomers (EV owners) and plan the corresponding ex-pected charging schedule for the EVs. The methods of esti-mating the energy requirements and setting up the chargingschedule may be different, but in general, the FOs try tominimize the charging cost of their customers as well asguarantee the driving requirements of the EV owners.

2) Market based approach for distribution grid congestionprevention—Offline scheduling. The distribution grid ca-pacity market will take effects if congestion happens. FOstrade the power capacity of the distribution grid in thismarket. During the negotiation of the market, a shadowprice will be issued by the market operator in the time slotwhere congestion happens. Then this shadow price will besent to FOs, FOs will add up the shadow price with the pre-dicted spot price and utilize the new price to set up a newcharging schedule. Again, the new schedule will be sentto the DSO/Market operator, such iteration will be termi-nated until the congestion is eliminated. After congestionmanagement, FOs bid the allowable power schedule to the

energy market, such as Nord Pool Spot market1 in NorthEurope.

3) Online scheduling and real time control. It is valuable forFOs to utilize the online scheduling stage and make bettercharging schemes, the general objectives are to avoid en-ergy imbalance and to optimally participate in the regu-lating power market. Usually, more accurate informationis provided to the FOs in this stage, FOs can judge whetherthey need to reschedule the charging plan based on theutility and risk analysis. With regard to real time control,one can assume that the EVs will charge according to theplan; however, if grid normal technical operation is com-promised, FO management can be overridden by the DSOoperation, such as using load shedding scheme.

4) Settlements. In this study, the settlements are carried outwith the final energy price, i.e., the sum of spot price andshadow price (Tax, transmission, distribution fees etc., arenot included here).

B. Operation Principle and Assumptions

1) FO/EVs operation• Aggregating EVs. From a practical perspective, it is as-sumed that EVs need to subscribe to one FO, maybe inthe form of signing a contract that is valid for certaintime period. Such subscriptionwould possibly followingthe existing geographical areas, i.e., the neighborhoodsupplied by one FO, under one substation. The mobilityof EV, in such context, will also require the roaming-re-lated agreement/standards among different FOs as wellas an standardized ICT infrastructure, in order to makesure the FOs can access the EV information immediatelywhen the EVs switch FOs.

• Predicting EVs driving pattern. We assume EVs havestandardized function of being able to be plug and play.In most cases, they will charge at home, which is sup-plied by their signed FO. By establishing a database forEV users, it is feasible for FOs to predict EVs drivingpattern. Besides, EV owners are encouraged to submittheir draft plan for the utilization of EVs in the nextday. In few cases, they may need to charge in someother areas where belong to another FO, since the fullycharged battery in morning time could sustain theirdaily driving requirements in most times. In such case,a roaming technology widely used in the Telecom couldbe an example for us, which means that FO and FO canshare information.

• Optimal charging schedule generation. We assume thatthe charging process of EVs following a linear behaviorand FOs use a linear programming technology to modelthe optimal charging problem of EV fleet and determinethe aggregated EV charging profiles. The computationspeed is quite fast for FOs.

• Interacting with DSO/Market operator. For the interac-tions between FOs and DSO/Market operator, it is as-sumed that ICT infrastructure can facilitate the commu-nication.

1http://www.nordpoolspot.com/



2) DSO/Market operator operation• Grid state estimation. The proposed solution requires ahigher level of automation on the operation of distribu-tion grid, such as demand forecasting, grid state estima-tion and online grid measurements. This is not an easytask for DSO at this moment, since little real time infor-mation exists about the power flow in the low voltagegrid. While in the medium voltage grid, the real timemeasurement is traceable in the Danish distribution sys-tems andmany other European distribution systems. Thegood news for the low voltage grid is the installation ofthe smart meters, today, in Denmark, half of all house-holds have an updated meter that can be read remotely atleast on hourly basis. With these inputs, the DSO couldstudy the co-variation between the DERs and traditionalload. Data mining technology can be used to separatethe information of conventional loads from EVs wouldbe highly necessary.

• Shadow price set up. The shadow price will be deter-mined based on the power requirement of FOs and thecapacity of transformer/line of the distribution grid.More details will be presented in next section.

III. METHOD DEVELOPMENT OF SMART CHARGING OF EVSWITH GRID CONGESTION MANAGEMENT

In this section, algorithms and models for enabling the EVcharging profile management are discussed. In short, FOs pre-dict their customers’ energy requirements and make the energyschedule, which is shown in Section III-A. Then Section III-Billustrates the method for grid congestion management. A set-tlement example is given in Section III-C. It is noted that themain work in this study is in the stage of offline scheduling,and it is assumed that there is no schedule changes in the onlinescheduling period and the EVswill charge based on the schedulemade in the offline scheduling stage.

A. Energy Schedule of the FOs Without CongestionManagement

We provide one solution as a reference for the energyschedule setting of FOs, the solution is based on our previousstudy [10], in which, an optimal charging strategy has beenproposed for EVs with the purpose of fulfilling EV owner’sdriving requirements as well as minimizing the charging cost.The solution is briefly modified and introduced as follows:

subject to

(1)

With the above optimization problem, the FO can generate aunique energy schedule for EV owner; the sum of the individualEV energy schedule will be denoted as , and

where

Number of EVs under FO .

Number of time slot in the scheduling period.

Number of FOs.

Index for the number of EVs under each FO,.

Index of time slot in the scheduling period,.

Index for the number of FOs, .

Predicted day-ahead electricity market pricevector.

Decision variable vector.

Length of each time slot.

Power requirements of EVs of each FO in eachtime slot.

Initial SOC of individual EV.

Recommended minimum SOC of the EV.

Edrive The predicted individual EV owners drivingrequirement.

Charge rate in term of energy of individual EV.

Recommended maximum SOC of the EV, whereis the parameter which express the charging

behavior of the battery of the EV is a linearprocess, is the capacity of the battery ofthe EV.

In (1), the first constraint means that the available energy inthe battery should be greater than or equal to the energy require-ment for the next trip. The second constraint indicates that theavailable energy in the battery should be less than or equal tothe power capacity of the battery. The third constraint repre-sents that the charging rate is less than or equal to its maximumpower rate of a charger. The physical meaning of the decisionvariable vector is to make a decision to distribute/chargethe power on the certain time slots, where the charging cost canbe minimized.

B. Market Based Approach for Distribution Grid CongestionPrevention

1) Analytical Analysis of Distribution Grid Capacity Market:In general, the method starts with a proposed cost functionwhich represents the cost of the power preference difference ofa FO in each time slot, e.g.,



To facilitate the understanding, we assume

(2)

where keep the same with above notation, denotes thecontrol variable, means the weighting factor which are as-sociated with the power difference, the larger implies asmaller difference.The objective is to minimize the cost functions as well as

respect to the constraint from DSO:

subject to

(3)

where is the power capacity specifically for all the FOs,for example, it can be estimated by the DSO after deducting theconventional loads.This problem is a convex optimization problem and relevant

research [29], [30] show that by introducing Lagrange multi-pliers or shadow price , problem (3) can be trans-ferred into following partial Lagrangian problem:

(4)

.The centralized optimization problem (3) is transferred into

a decentralized one with associated shadow price in eachtime slot, with the purpose of emulating the market behavior.Following work aims to find the optimal power for each FO

and the associated shadow price on the distribution grid line.We will assume, for simplicity, the Lagrangian function has aunique minimizer over , which denoted as .The dual problem (4) is then given by

(5)

This dual problem (5) will be solved by projected subgradientmethod [31]–[33],

(6)

where is the subdifferential of at and, one can find

with being the solution to the following opti-mization problem:

(7)

Fig. 3. Flow chart of the proposed cost and schedule adjustment algorithm.

This optimization is completely separable between various FOs,and can therefore be solved distributively. For each FO, such as

, the optimization problem becomes

(8)

Solving problem (8) for the FOs gives power that canbe used to find the subgradient:

(9)

2) Cost and Schedule Adjustment Algorithm: The followingsteps illustrate the cost adjustment algorithm which are illus-trated in Fig. 3 and can mimic the trading and negotiationprocess in the distribution grid capacity market, when con-gestion happens. The algorithms integrate the mechanismsdiscussed in the above of this section.1) FOs submit their energy schedule to the distribution gridcapacity market before submitting them to the electricityspot market.

2) The DSO/Market operator predicts whether congestionwill happens based on the schedules of FOs, if happens,go to the distribution grid capacity market, otherwise, theenergy schedule is approved.

3) Distribution grid capacity market operation



Fig. 4. Spot price in one day (12, Jan, 2011), DK-West, from NordPool, whichis used as the predicted price for fleet operator.

Initialize dual variable , e.g., usingor .

loop• With the parameter of , market operator determinesthe capacity margins using (9) based on the solu-tions of (8).

• Update the variables , untilthe prices converge.

4) The new shadow price will be sent to FOs, FOs will addup this price on top of predicted spot price, recalculatethe optimization problem (1), and get a new energy/powerschedule and send to the DSO/Market Operator.

5) Do the above calculation (step 2 and 3) again, until, then terminate the iteration.

6) Bid final energy/power schedule to the electricity spotmarket.

In above algorithm [31], denotes the step size and canbe chosen as which is a positive constant, independentof ; with such choice, the convergence is guaranteed.

C. Settlements

In the settlement stage, the sum of the electricity spot priceand shadow price will be used as an energy price, and the cor-responding cost for FOs are given by

(10)

IV. CASE STUDY

In this case study, a representative distribution grid is illus-trated in Fig. 2. It is assumed that 60 households are connectedon the feeder. Sixty percent of the consumers are assumed tohave EVs which are operated by and . Specifically,

represents the capacity of the transformer/cable and thiscapacity will be shared by and during the schedulingand operation period.

Fig. 5. Top: Aggregated energy demand of and ; Bottom: Aggre-gated energy schedule of and .

Fig. 6. A hypothetical cost function of (2), has a stepwise increase of 0.1from 0.1 to 1.

A. Energy Schedule of the Fos Without CongestionManagement

For the EV charging schedule, the information of hourly elec-tricity spot price of the Nordic power market is assumed to beperfectly known by the FOs, and the price data is identical withprevious study [10], which is illustrated in Fig. 4. The artificialdriving data of the EV fleet have been generated based on the2003AKTASurvey [34], in which 360 cars in Copenhagenweretracked using GPS from 14 to 100 days. Each data file includesstarting and finishing time, and the corresponding duration anddistance. The original data is transferred into 15 minutes in-terval driving energy requirements based on the assumption of11 kWh/100 km. The energy requirements of and arethe sum of the 18 EVs, which is illustrated in the top of Fig. 5. Itmay not be easy to identify the individual EV’s driving energyrequirements, but a general trend can be concluded that mostof the driving time is located in the morning and evening pe-riods. In , EV12 has the largest energy requirement whichis 15.45 kWh. In , it is EV4 that needs the most energy



Fig. 7. Left: The sum of the spot price and shadow price in each iteration step. Right: The comparisons of FO’s power schedule in each step with the powercapacity.

which is 11.55 kWh. We assume that the data used for the simu-lation which represent the EV owners’ driving requirements areperfectly known to the FOs. The 15 minutes interval is change-able rather than absolute. For other parameters:• Battery capacity is set to 20 kWh• is set to 0.2 of the battery capacity• is set to 0.2 of the battery capacity• is set to 0.85 of the battery capacity• Maximum charging power is limited to 2.3 kW, this fitswith the Danish case (10 A, 230 V connection).

With this information, one obtains the aggregated chargingenergy of each FO, which is shown at the bottom of Fig. 5.It is observed that the charging period is concentrated on theearly morning time due to the lower electricity price in that timeperiod. We assume that the power is constant in one time slot,which means the corresponding power in each time slot can beobtained (Energy/ ).

B. Market Based Approach for Distribution Grid CongestionPrevention

In this step, we will illustrate the effectiveness of utilizingthe shadow price, i.e., to facilitate the congestion manage-ment in the proposed method. It is noted that the cost functionin this study presented by the quadratic function is assumed torepresent the cost for the energy preference loss. The accuracyof the cost function is out of the scope of this study, the focusis to show how the FOs establish the schedule based on the costfunction and the shadow price. Fig. 6 illustrates the cost func-tion of (2) with various , in which is set to 30 kW.The power capacity is set up according to the trend in

the real case; generally, the capacity is higher in the later eveningand early morning time and lower in the day and evening time.The curve of the power capacity is shown by the surface in theright figure of Fig. 7. The weighting factor rate is setto 0.5 and 0.1. The value of is chosen as 0.1 in this case.Note that the variable and are connected, an appropriatevalue of the two variables can ensure smooth operation of theproposed method, i.e., the trade-off of the speed of the conver-gence and the accuracy of the solution. However, there is not astrict rule for choosing the parameter values. Together with the

energy schedule of and before congestion manage-ment, the values of power capacity and weighting factor rate,the simulations are presented in the Fig. 7. From these two fig-ures, it cab be seen that the congestion problems are solved after5 steps. Note that in the beginning, the shadow price is zero, sothe blue curve represents the same price information as the onein Fig. 4. The purpose for put this price again is to get a com-plete view on the change of the price. Same explanation holdsfor the blue power curve in the right figure.Fig. 7 presents the dynamic process of the distribution grid

congestion management. It is noted that in each iteration step,the negotiation process of the FOs in the distribution grid ca-pacity market is not shown, i.e., only the final shadow priceis presented. But in order to see the effectiveness of the distri-bution grid capacity market, Fig. 8 is presented. In this figure,one can note the convergent process of the shadow price inthe second iteration of Fig. 7. During the time slot of 9 to 16,the total power demands from and are same, but thepower capacity various from 70 kW to 56 kW with a stepwisedecrease of 2 kW. The result shows that the lower power ca-pacity results in higher shadow price. Besides, the steady stateis reached quickly.

C. Day-Ahead Congestion Settlements

The charging cost of and are compared from twotime periods, one is the cost before congestion management,and another is the cost after congestion management. Table Ipresents the results which show that charging cost of each FOincrease a lot. It indicates a shortage of the distribution capacity.

V. DISCUSSION AND CONCLUSION

In this paper, two control issues are integrated in a lowvoltage active distribution system consisting of three actors,DSO/Market operator, FO and EV owner. One is the optimalcharging of EVs, another is the congestion management onthe distribution system level. Two steps are adopted to addressthese two issues, linear programming is firstly used to modelthe charging process of EVs and to produce an aggregatedenergy schedule of FOs. If the sums of the energy schedule of



Fig. 8. Convergence of toward the shadow price inthe second iteration of Fig. 7, the converged price in each time slot of this figurecorresponds to the circled values in the Fig. 7(-to left).

TABLE ICHARGING COST OF FOS IN THE PERIOD OF BEFORE AND AFTER CONGESTION

MANAGEMENT

the FOs overload the distribution grid, then, a distribution gridcapacity market scheme is adopted to coordinate the energyschedule. The proposed solution for solving the congestionproblem and managing the charging of EVs is an integration ofa direct control and a price-based coordination. It is believedthat the safety operation of the distribution grid can be highlyensured by coordinating the relation between the three marketactors with the proposed framework. We also expect that suchcoordination strategy can be used to control other smart appli-ances, including thermostatically controlled loads such as heatpumps. As we discussed before, the market scheme can also beused in the real time for congestion relief.We want to point out that voltage control is also an impor-

tant issue for distribution grid operation, although we did notconsider it in this study. In a practical way, in the planning pe-riod, DSO considers and pre-handles it by reinforcing the gridinfrastructure based on the regulations already existed whichdescribe the allowed voltage safety bands in the distributiongrid. In the normal operation period, in the substation level,transformer has a tap change which can be used to regulatethe voltage. Such as in Denmark, in general, 60 KV/10 KVtransformer is an OLTC (On-load tap-changers). In the future,system operator could set up some grid codes for DERs, re-quiring the DERs to have their own embedded voltage control,which could solve the problem preventively. In the context ofthis study, voltage control can also be implemented by marketscheme. This approach would be to establish a market wherevoltage stabilizing services can be bought or sold. Technically,it is feasible. However, it is not easy to identify and validatethe committed power from the DERs, which will bring manychallenges. Besides, the AC power flow calculation will alsointroduce time-consuming problem to this method. In a simple

market way, due to the increasing penetration of distributed gen-eration, DSO can solve this problem by a contract-based solu-tion, such as DSO can sign some contracts with FOs to get therequired services.It is noted that market approach has been well discussed and

considered as one of the best approach for resource allocation,meanwhile, we also see some practical point deserve discussionfor utilizing this approach for congestion management, mainlyfrom three perspectives: 1) Stakeholder’s acceptance on usingprice coordination: Although using price coordination approachwould possibly enable an optimal resource allocation, but theuncertain in terms of end-user involvement, clear businessmodels for FOs and DSOs, and the necessity for the regulatorysupport makes using price as a coordination tool for servinggrid services a challenging task. 2) Size of the market and theassociated market power issues: To ensure enough competitionand fairness of the capacity market, one prerequisite is thenumber of market participants, i.e., FOs. If there are few FOsin the distribution area, issue e.g., market power will becomea major challenge from the market perspective. 3) The infor-mation communicating between various stakeholders and thesupporting ICT infrastructure: FOs need to communicate wellwith EVs in order to make an optimal charging schedule, theseinformation include driving pattern, state of charge, some otherpreferences of EV owners. The time-stringent is not an essentialissue here, however, EV owners cooperation are much wanted.For the interaction between FOs and distribution grid capacitymarket operator, real time communication is a challenge, butwe can set up a reasonable time range and also try to limitingthe market iteration with certain rules. This kind of setup willrequire advanced ICT infrastructure.To sum up, the proposed method is flexible and scalable and

can technically be enhanced to provide a complete set up for thecongestion prevention in the scheduling period and congestionrelief in real time, by taking into account the discussion above.Additional, the practical points discussed above imply the eco-nomic feasibility should also be analyzed in future.

ACKNOWLEDGMENT

The authors would like to thank our partners of WP4, iPowerproject for the discussion and inspiration. Besides, the authorsare grateful to their colleagues K. Heussen and P. B. Andersenfor the discussion on the grid congestion management. Also, theauthors are grateful to the financial support of the Danish iPowerproject.

REFERENCES[1] F. Tuffner andM. Kintner-Meyer, “Using electric vehicles to meet bal-

ancing requirements associated with wind power,” Pacific NorthwestNational Laboratory (PNNL), Richland, WA, USA, Tech. Rep., 2011.

[2] J. Pecas Lopes, P. Rocha Almeida, and F. Soares, “Using vehicle-to-grid to maximize the integration of intermittent renewable energy re-sources in islanded electric grids,” in Proc. 2009 IEEE Int. Conf. CleanElectr. Power, pp. 290–295.

[3] J. Lopes, F. Soares, P. Almeida, and M. da Silva, “Smart chargingstrategies for electric vehicles: Enhancing grid performance and maxi-mizing the use of variable renewable energy resources,” in Proc. 24thInt. Battery, Hybrid, Fuel Cell Elect. Veh. Symp. Exhib. (EVS24), Sta-vanger, Norway, 2009.

[4] W. Kempton and J. Tomic, “Vehicle-to-grid power fundamentals: Cal-culating capacity and net revenue,” J. Power Sources, vol. 144, no. 1,pp. 268–279, 2005.



[5] W. Kempton and J. Tomic, “Vehicle-to-grid power implementation:From stabilizing the grid to supporting large-scale renewable energy,”J. Power Sources, vol. 144, no. 1, pp. 280–294, 2005.

[6] N. Rotering andM. Ilic, “Optimal charge control of plug-in hybrid elec-tric vehicles in deregulated electricity markets,” IEEE Trans. PowerSyst., vol. 26, no. 3, pp. 1021–1029, Aug. 2010.

[7] S. Han, S. Han, and K. Sezaki, “Development of an optimal vehicle-togrid aggregator for frequency regulation,” IEEE Trans. Smart Grid,vol. 1, no. 1, pp. 65–72, 2010.

[8] O. Sundström and C. Binding, “Optimization methods to plan thecharging of electric vehicle fleets,” in Proc. Int. Conf. Control,Commun., Power Eng., pp. 28–29.

[9] T. Kristoffersen, K. Capion, and P. Meibom, “Optimal charging ofelectric drive vehicles in a market environment,” Appl. Energy, 2010.

[10] J. Hu, S. You, J. Oestergaard, M. Lind, and Q. Wu, “Optimal chargingschedule of an electric vehicle fleet,” in Proc. 2011 46th Int. Universi-ties’ Power Eng. Conf. (UPEC), pp. 1–6.

[11] T. Lan, J. Hu, G. Wu, S. You, L. Wang, and Q. Wu, “Optimal chargecontrol of electric vehicles in electricity markets,” in Part of: Proc. 4thInt. Workshop Intell. Inf. Manage. Syst. Technol., 2011.

[12] C. Sndergren, “Edison project: Electric vehicles in future marketmodels,” Danish Energy Association, Tech. Rep., 2011.

[13] G. Heydt, “The impact of electric vehicle deployment on load man-agement strategies,” IEEE Trans. Power App. Syst., vol. 102, no. 5, pp.1253–1259, 2007.

[14] R. Green, II, L. Wang, and M. Alam, “The impact of plug-in hybridelectric vehicles on distribution networks: A review and outlook,” Re-newable Sustain. Energy Rev., vol. 15, no. 1, pp. 544–553, 2011.

[15] J. Lopes, F. Soares, and P. Almeida, “Identifying management proce-dures to deal with connection of electric vehicles in the grid,” Proc.2009 IEEE Bucharest PowerTech, pp. 1–8.

[16] K. Clement-Nyns, E. Haesen, and J. Driesen, “The impact of chargingplug-in hybrid electric vehicles on a residential distribution grid,” IEEETrans. Power Syst., vol. 25, no. 1, pp. 371–380, 2010.

[17] O. Sundstrom and C. Binding, “Planning electric-drive vehiclecharging under constrained grid conditions,” in Proc. 2010 IEEE Int.Conf. Power Syst. Technol. (POWERCON), pp. 1–6.

[18] O. Sundstrom and C. Binding, “Flexible charging optimization forelectric vehicles considering distribution grid constraints,” IEEETrans. Smart Grid, vol. 3, no. 1, pp. 26–37, 2012.

[19] J. Lopes, F. Soares, and P. Almeida, “Integration of electric vehiclesin the electric power system,” Proc. IEEE, vol. 99, no. 1, pp. 168–183,2011.

[20] P. B. Andersen, J. Hu, and K. Heussen, “Coordination strategies fordistribution grid congestion management in a multi-actor, multi-objec-tive setting,” in Proc. 2012 Eur. ISGT, Berlin, Germany.

[21] R. D. Zimmerman, “Uniform price auctions and optimal power flow,”Matpower Technical Note, Feb. 1, 2010 [Online]. Available: http://www.pserc.cornell.edu/matpower/TN1-OPF-Auctions.pdfF

[22] B. Biegel, P. Andersen, J. Stoustrup, and J. Bendtsen, “Congestionmanagement in a smart grid via shadow prices,” in Proc. 8th IFACSymp. Power Plant Power Syst. Control, Toulouse, France, Sep. 2012.

[23] N. O’Connel, Q. Wu, J. Stergaard, A. Nielsen, S. Cha, and Y. Ding,“Electric vehicle (ev) charging management with dynamic distributionsystem tariff,” in Proc. 2011 ISGT.

[24] A. Mohsenian-Rad, V. Wong, J. Jatskevich, R. Schober, and A. Leon-Garcia, “Autonomous demand-side management based on gametheo-retic energy consumption scheduling for the future smart grid,” IEEETrans. Smart Grid, vol. 1, no. 3, pp. 320–331, 2010.

[25] Z. Ma, D. Callaway, and I. Hiskens, “Decentralized charging controlfor large populations of plug-in electric vehicles,” IEEE Trans. ControlSyst. Technol., vol. 21, no. 1, pp. 67–78, 2013.

[26] A. Ipakchi and F. Albuyeh, “Grid of the future,” IEEE Power EnergyMag., vol. 7, no. 2, pp. 52–62, 2009.

[27] C. B. Lotte, H. Rasmussen, and M. Togeby, “Managing congestion indistribution grids-market design consideration,” Ea Energy Analyses,Tech. Rep., 2012.

[28] M. D. Galus, C. Dobler, R. A. Waraich, G. Andersson, R. A. Waraich,R. A. Waraich, G. Andersson, and G. Andersson, “Predictive, dis-tributed, hierarchical charging control of PHEVs in the distributionsystem of a large urban area incorporating a multi agent transportationsimulation,” ETH, Eidgenössische Technische Hochschule Zürich,IVT, Institut für Verkehrsplanung und Transportsysteme, 2011.

[29] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge,U.K.: Cambridge Univ. Press, 2004.

[30] S. Boyd, L. Xiao, A. Mutapcic, and J. Mattingley, “Notes on decom-position methods,” Notes for EE364B, Stanford Univ., 2007.

[31] S. Boyd, L. Xiao, and A. Mutapcic, “Subgradient methods,” Lecturenotes of EE392o, Stanford Univ., Autumn Quarter, vol. 2004, 2003.

[32] M. Grant, S. Boyd, and Y. Ye, “CVX users guide,” Tech. Rep., Tech-nical Report Build 711, 2009 [Online]. Available: http://citeseerx.ist.psu.edu/viewdoc/download

[33] M. Grant and S. Boyd, “Cvx: Matlab software for disciplined convexprogramming,” 2008 [Online]. Available: http://cvxr.com/cvx/

[34] L. Christensen, “Edison Project Report: Wp1.3 Electric vehicles andthe customers,” Tech. Rep., DTU Transport, 2011.

Junjie Hu received the Master of Engineering incontrol theory and control engineering from TongJiUniversity, Shanghai, China, in 2010. Currently, heis a Ph.D. student at the Department of ElectricalEngineering within the Technical University ofDenmark (DTU). His Ph.D. project focuses onintegrating control policies on controllable load,mainly electric vehicle, for active power distributionsystem, of which the control policies are directcontrol and indirect (price) control.

Shi You received the M.Sc. degree in electricalengineering from Chalmers Institute of Technology,Sweden, and the Ph.D. degree in electrical engi-neering from Technical University of Denmark,in 2006 and 2011, respectively. He is currently aPostdoctoral Researcher with the Department ofElectrical Engineering, Technical University ofDenmark. His main fields of interest are renewableenergy integration, distribution grid planning, andelectricity markets.

Morten Lind is Professor Emeritus of Automationat Department of Electrical Engineering at TechnicalUniversity of Denmark and is associated with theCentre for Electric Power and Energy. His researchinterests include automation design, supervisorycontrol of complex industrial systems and infra-structures, functional modeling, and application ofagent technology and knowledge based system inautomation.

Jacob Østergaard (M’95–SM’09) received theM.Sc. degree in electrical engineering from theTechnical University of Denmark (DTU), Lyngby,Denmark (DTU), in 1995.He is a Professor and Head of the Centre for Elec-

tric Power and Energy, Department of Electrical En-gineering at DTU. From 1995 to 2005, he worked forthe Research Institute for the Danish Electric Utili-ties. His research interests include system integrationof wind power, control architecture for future powersystems, and demand side.

Prof. Østergaard is serving in several professional organizations including theEU SmartGrids advisory council.

Multilevel coordination in smart grids for congestion management ofdistribution grid 145

A.7 Multilevel coordination in smart grids forcongestion management of distribution grid

This paper was presented at International conference on intelligent system ap-plications to power systems (ISAP), 2013, Tokyo, Japan.

Multilevel Coordination in Smart Grids forCongestion Management of Distribution Grid

Junjie Hu, Morten LindCenter for Electric Power and Energy


Email: {junhu, mli }@elektro.dtu.dk

Arshad SaleemSchool of Electrical Engineering

Royal Institute of TechnologySE-100 44 Stockholm, Sweden

Email: [email protected]

Shi You, Jacob ØstergaardCenter for Electric Power and Energy


Email: { sy, joe }@elektro.dtu.dk

Abstract—The operation of the distribution network willchange in the near future due to increasing size and number ofdistributed energy resources (DER) and demand side resources(DSR). An active distribution network is proposed to addressthe challenges. The normal operation of an active distributionnetwork requires coordination of different values and operationconstraints of various involved actors. This paper proposes amultilevel coordination strategy for congestion management ofdistribution network. Firstly, the scheme of an active distributionnetwork is described. Then, the coordination strategies betweenvarious actors, i.e., distribution system operator (DSO), fleetoperators (FO), and EV owners are discussed. Further, a mathe-matical formulation of the chosen coordination strategies betweenDSO and FOs are presented and some case studies are shownto illustrate the effectiveness of the proposed solutions. Finally,we give the argument and proposal of using multi-agent basedplatform to demonstrate the multilevel coordination solution.

Index Terms—Congestion management, Distribution grid,Multilevel coordination, Multiagent systems based platform

I. INTRODUCTION

Denmark was a pioneer in wind power which provides alarge amount of electricity to Danish consumers, at the endof 2012 [1], the total installed wind capacity in the Danishpower gird was 4,162 MW which share 30% of the domesticelectricity usage. Wind energy in Denmark is expected to growdue to the political strategy of 50% wind power in the 2020Danish power system [2]. The total installed wind power inDenmark is connected at the distribution system level, whichbring challenges to Energinet (TSO of Denmark). Energinethas limited or no access to the information about the state atthe low grid voltage level. In order to address the challenges,several actions [3] have been implemented or planned, suchas

• Coordinate the power flows among different systems byelectrical interconnections, mostly high voltage directcurrent to the TSOs in the Sweden, Norway, Germany,and soon the Netherland.

• Balance the power system by the deregulated power mar-ket with the collaborations of power Balance ResponsibleParties (BRP). The BRPs make the power and energy bidsinto the market, consists of conventional power and windpower.

• Implement a tool that provides real time estimation of theamount of injections from wind energy.

• Manage the flexible demand, like electric vehicles, heatpumps.

With the expected development towards a power systemdepentent on intermittent renewable energy sources, the needfor some of the ancillary services is likely to increase, espe-cially for balancing purposes. Both EVs and heat pumps arebelieved to play important roles in balancing the system. Inorder to aggregate the flexibilities of demand and capturingthe business opportunities of providing the service to thesystem operator, a new business entity, namely fleet operator(FO) has recently emerged [4], [5]. Alternative names for anFO are virtual power plant (VPP), aggregator or chargingservice provider. However, the operation of the distributiongrid may be challenged due to the increasing size and numberof consumption units which can cause problems in peakhours. Besides, there exists facts that the closer the renewableproduction installed to the consumer premises and the con-sumer’s awareness of consumption. As a result, the DSO havestarted to recognize the necessity for electricity distributionand operation evolving from the usual passive unidirectionalflow network to an active distribution network [6].

In this paper, we consider a particular case combiningEV charging cost minimization and distribution grid capacitymanagement (active power transfer capacity). Previous studies[7]–[9] has shown that EVs can provide valuable servicesto the system operator, for example, during strong windconditions, where the total wind power production capabilitybecomes highly utilized, the need of maintaining balance be-tween production and consumption might increase and can beprovided by utilizing the controllable flexibilities of EVs. As aconsequence the distribution system might become overloaded.Besides, the spot electricity price might become cheap whenthe wind power penetration is high. This will also furtherincrease the consumption on the distribution grid side. In orderto address the challenge, our study aims to answer how thevalues and operation constraints among the DSO, FOs and EVscan be coordinated within a market based platform sitting inan active distribution network.

The remainder of the paper is organized as follows: Insection II, a general introduction regarding the operation ofdistribution network today and an active distribution networkis given. Besides, the scheme of an active distribution network

is also presented. Section III mainly presents the coordinationstrategies between DSO and FOs, FOs and EV owners. Then aframework for design and method development for multilevelcoordination is illustrated in section IV. A case study is givenin section V to demonstrate the proposed method. Further, theproposal of using Multi-Agent Based Platform to demonstratethe multilevel active distribution systems are made in sectionVI. Finally, discussion and conclusions are made in sectionVII.

II. MULTI ACTOR SETTING, MULTILEVEL COORDINATIONIN AN ACTIVE DISTRIBUTION NETWORK

A. Distribution network operation

1) Distribution Network Operation Today: DSO tasks inconventional system operation [10], are mostly focused on‘off-line’ tasks related to asset management and maintenanceduring normal system conditions. The primary objective underemergency conditions is to organize restoration of the networkas quickly as possible. Distribution systems today tend to beweakly monitored as compared to transmission grids, and con-trolled in a decentralized fashion on the basis of preconfiguredlocal controls (e.g. by means of grid codes and protectionsettings).

The key operations of the DSO are:• Grid dimensioning (incl. contigency planning and load

curve estimation)• Maintenance and outage related topology reconfiguration• Adjustment of transformer taps• Fuses and relay operation• Fault-analysis and repair• Logging events and standard management report• Managing trouble call information and inform customers.2) Operation in an active distribution network: To illustrate

a future operation scenario with a higher level of automation,it is considered how the above operations can be extendedwith additional online- and data intensive acquisition. Inorder to identify and solve congestion problems, the DSOrequires additional measurement equipment and/or technologyenabling identification and anticipation of load patterns andgrid ‘bottlenecks’.

Key Operations for DSO congestion management in anactive distribution network would be:

• Demand forecasting• Grid state estimation• Online grid measurements• Real-time intervention in case of unexpected deviations

challenging grid reliability• Meter data collection and aggregation

B. The scheme of an active distribution network with multi-actors, multilevel coordination

Fig. 1 shows the scheme of an active distribution network,in which four types of actors are loaded on different levels.In general, each of the actors is associated with a kind ofoperations, namely, the DSO is responsible for the reliability

FO1

Fleet Operation

Electricity

Spot Market

Grid

Measurements

DSO

Distribution Operation

Conventional

load

FOn

Fleet Operation

Market Based

Platform for

Distribution Grid

Congestion

Management

Information

flow

Control

relation

Physical

connection

Regulation

Market

Fig. 1. Actors (stakeholders), problem domain and main information andcontrol flows within an active distribution network.

of the distribution network, FOs are responsible for makingthe energy schedules, biding it into traditional market such asday-ahead spot market and regulation market and providingthe electricity for end users, EV owners are taking care ofthe charging of their EVs by subscribing to an FO or makingthe charging decision by themselves. By introducing a marketbased platform on the distribution grid level, the DSO willcoordinate their requirements with market operator who theninteract with FOs. About the control/coordination relationsbetween FOs and EVs, this could be implemented either indirect control or indirect control method. Further discussionregarding the multilevel coordinations will be presented in nextsection.

III. COORDINATION METHOD AMONG THE VARIOUSLEVELS OF THE ACTIVE DISTRIBUTION SYSTEMS

A. Coordination method between DSO and FOs

Generally, the market based platform will be used to coordi-nate the requirements and values of DSO and FOs. The mech-anism behind the market could be designed in many ways,such as various potential tariff regimes [11] [12], uniform priceauction mechanism [13], shadow price based mechanism [14].We briefly introduce three types of mechanisms in the below:

1) Dynamical grid tariff: In this method [12], the DSOgenerates a time and grid-location dependent price for gridusage based on expected nodal consumption levels. The DSOanticipates the size and the price-responsiveness of the loadat critical grid nodes and calculates the price to optimallyreflect the expected congestion problem. The FO will then getthe dynamic nodal tariff and make an optimal schedule withrespect to the predicted spot price and dynamic grid tariff.

2) Uniform price auction mechanism : The uniform priceauction [13] can be designed as either single-sided auctionsor two sided auctions. This will fully depends on the scale ofthe market, i.e., whether it is used by single DSO or multiDSOs supposing there are several FOs. It is noted that theuniform price auction mechanism is usually combined with

optimal power flow calculation, which mean either marketoperator/DSO will implement a lot of calculations.

3) Shadow price based mechanism: In this method [14],FOs will submit power requests to DSO for their aggregatedenergy/power schedule on each node (aggregated capacity) be-fore submitting the energy schedule to the day-ahead market;in response they will receive a price for each node whichreflects the respective congestion, and are requested to updatetheir energy schedules. The process will terminate when allconstraints are satisfied.

B. Coordination method between FOs and EV ownersResearch in [15], [16] give a comprehensive review on

the control strategies for flexibility aggregation. Three controlarchitectures are examined and compared in [15], namely cen-tralized load control, hierarchical load control via aggregatorsand distributed control. The control method is also discussedin [15], in which direct control and indirect control in the formof price signal are described. Direct control means that FO candirect schedule and control the charging of EVs [17]. Indirectcontrol implies that FO coordinate the charging of EVs byeither two way [18], [19] or one side price signals [20]. EVowner determine the charging profile of EVs by themselves. Ashort comparison between direct control and indirect controlis given in Fig. 2.

Control polices

Direct control Indirect control

· Control signals

· High level controller

make the decision

· Price incentive

· Consumers make decisionFeatures of the

control policies

Advantages

Disadvantages

· High certainty

· Better optimal results

· Privacy improved

· Less communication cost

· Sophisticated hardware

· High communication

cost

· Lower certainty

· Demand Price elasticity is

required

Fig. 2. Comparison between direct control and indirect control strategy

IV. FRAMEWORK DESIGN AND METHOD DEVELOPMENTFOR MULTILEVEL COORDINATION

Within the assumed system architecture, we propose aframework consisting of four fundamental stages [21], i.e.,offline scheduling, online scheduling, real time control andsettlement, to operate and control this system. These principlesare in line with the system control function required at thecontrol center of power system operation [10], i.e., instan-taneous operation, operation planning and operation report-ing. This framework is a fully charging profile managementconsidering charging cost minimization and distribution gridcapacity management. During the framework, shadow pricebased mechanism designed for the market and direct controlbetween FOs and EV charging are used.

A. Framework design and method development for multilevelcoordination

1) Energy schedule of the FOs without congestionmanagement-Offline schedulingAll the FOs need to predict the energy requirements(driving patterns) of their customers (EV owners) andplan the corresponding expected charging schedule for theEVs. The methods of estimating the energy requirementsand setting up the charging schedule may be different,but in general, the FOs try to minimize the chargingcost of their customers as well as guarantee the drivingrequirements of the EV owners.

2) Market based approach for distribution grid congestionmanagement-Offline schedulingThe market based platform will be used if congestionhappens and the shadow price based mechanism is chosenfor the market operation. FOs trade the power capacity ofthe distribution grid in this market. During the negotiationof the market, a shadow price will be issued by the marketoperator in the time slot where congestion happens. Thenthis shadow price will be sent to FOs, FOs will send backa new schedule to the Market operator, such iteration willbe terminated until the congestion is eliminated.

3) Online scheduling and Real time controlIt is valuable for FOs to utilize the online scheduling stageand make better charging schemes, especially regardingthe participation in the regulating power market. Besides,if more accurate information is provided to the FOs, FOscan judge whether they need to reschedule the chargingplan during this stage. With regard to real time control,one can assume that the EVs will charge according tothe plan; however, if grid normal technical operation iscompromised, FO management can be overridden by theDSO operation, such as using load shedding scheme.

4) Settlements The settlements need to be designed wellconsidering both the spot price and shadow prices. Be-sides, tax, transmission and distribution fees etc., shouldbe taken into account.

B. Market based approach for distribution grid congestionmanagement

In this subsection, we mainly focus on introducing theshadow price, where it comes from, how it can be utilizedin the study.

1) Analytical analysis of shadow price based market op-eration: In general, the method starts with a proposed costfunction which represents the cost of the power preferencedifference of a FO in each time slot, e.g.,

µk = ζk(Pk,i).


µk = Ck,i(Pk,i − PEk,i)

2, (1)

where k, i denote the index for the number of FOs and timeslot in the scheduling period, k = 1, ..., NB , i = 1, ..., NT

PEk,i means the schedule planned by FOs, Pk,i denotes the

control variable, Ck,i means the weighting factor which areassociated with the power difference, the larger Ck,i impliesa smaller difference.

The objective is to minimize the cost functions as well asrespect to the constraint from DSO:

minimize

NB∑

k=1

NT∑

i=1

Ck,i(Pk,i − PEk,i)

2

subject toNB∑

k=1

Pk,i ≤ PCap(i), i = 1, ..., NT , (2)

where PCap(i) is the power capacity specifically for all theFOs, for example, it can be estimated by the DSO afterdeducting the conventional loads.

This problem is a convex optimization problem and rele-vant research [22], [23] show that by introducing Lagrangemultipliers or shadow price Λ(i) ∈ RNT , problem (2) can betransferred into following partial Lagrangian problem:

L =

NB∑

k=1

NT∑

i=1

Ck,i(Pk,i−PEk,i)

2+

NT∑

i=1

Λ(i)(

NB∑

k=1

Pk,i−PCap(i))

(3).

The centralized optimization problem (2) is transferred intoa decentralized one with associated shadow price Λ(i) ineach time slot, with the purpose of emulating the marketbehavior. In our previous study [24], a strict mathematicalproof is presented for the justification of the shadow pricebased mechanism.

2) Cost and schedule adjustment algorithm within the mar-ket based platform: The following steps illustrate the interac-tions among DSO, FOs, and market operator, cost adjustmentalgorithm and can mimic the trading and negotiation processin the market operation, when congestion happens.

(1) FOs submit their energy schedule to the DSO beforesubmitting them to the electricity spot market.

(2) The DSO predicts whether congestion will happensbased on the schedules of FOs, if happens, FOs need togo to the capacity market, otherwise, the energy schedule isapproved.

(3) Capacity market operation• FOs send their power schedule PE

k,i to market operator.• Market operator determines the shadow price Λ(i) and

sends the price to FOs.• FOs update their power schedule according to the shadow

price and send it again to market operator.• Such iteration will be terminated according to certain

criteria, e.g., price convergence.Intuitively, Fig. 3 illustrate the operation sequence.

V. NUMERICAL EXAMPLES

In this step, we will illustrate the effectiveness of utiliz-ing the shadow price, i.e., Λ(i) to facilitate the congestion

FOs modify and send

the energy schedule to

the market operator

After serveral interations,

the process will converge

Market

operator

Fleet

operators

FOs submit their

energy schedule to

market operator.

Accepted by all

FOs.

Market operator will

issue a new price.

market operator initial a

price regarding the

utilization of the grid

Market operator

finally publish a price

Fig. 3. Convergence of Λ(i), i = 9, 10, ..., 16 toward the shadow price.

management in the proposed method. It is noted that the costfunction in this study presented by the quadratic function isassumed to represent the cost for the energy preference loss.The accuracy of the cost function is out of the scope of thisstudy, the focus is to show how the FOs establish the schedulebased on the cost function and the shadow price.

The weighting factor rate C1,i, C2,i is set to 0.5 and 0.1.The value of αω is chosen as 0.1 in this case. Note that thevariable Ck,i and αω are connected, an appropriate value ofthe two variables can ensure smooth operation of the proposedmethod, i.e., the trade-off of the speed of the convergence andthe accuracy of the solution. However, there is not a strict rulefor choosing the parameter values. The power capacity Pcap(i)is set up according to the trend in the real case; generally, thecapacity is higher in the later evening and early morning timeand lower in the day and evening time. Fig. 4 is presentedto note the convergent process of the shadow price. Duringthe time slot of 9 to 16 (15 minutes based time slot in a 24hours time window), there exists congestion in the network,the total power demands from FO1 and FO2 are same inthese time slot, (i.e., 39.1 kw from FO1 and 41.4 kw fromFO2), but the capacity reserved for these time slot is 70kW.The result shows that the steady state is reached quickly. Notethat each FO will obtain the final power schedule after themarket operation. Because of page limitation, FO1’s exampleis shown here, as presented in Fig. 5. From Fig. 5, one cansee that the newly obtained power, i.e., the green curve isquite close to the blue curve, this is because FO1 have higherweighting factor rate, which further imply that FO2 will needto reduce a little more power comparing to FO1.

VI. GRID CONGESTION MANAGEMENT WITHIN AMULTI-AGENT SYSTEMS BASED PLATFORM

In the discussions above, it is observed that some generaldesign principles are used such as decomposition, abstractionand scalability. These principles match the inherent capabilitiesof software agents and multiagent systems. In fact, multiagentsystem have been widely considered for control of power sys-tems [25], starting from a low level control of devices to higherlevel of planning and optimization. A detailed explanationfor the application of the three principles mentioned above



is presented below:• Decomposition: The electricity supply for the ender users

is provided by several FOs. The congestion problems ina distribution network can be decomposed into subprob-lems, which the different DSO may face challenges ondifferent levels of their grid. Some DSO might foreseeproblems on the medium voltage grid, while others mayencounter potential problems with capacity in the lowvoltage transformers.

• Abstraction: Abstraction can be used to define a simpli-fied model of the system that emphasizes some of thedetails or and suppresses others and to organize networkoperation. In this study, for example, FOs can be usedto abstract the requirements and operations constraint ofEV owners. In addition, the radial distribution grid canbe abstracted in the form of Fig. 1 when emphasizing theactive power transfer capacity management.

• Scalability: Multiagent systems have mature mechanismfor implementation of cooperative and competitive mech-anisms. These mechanism can be used in the interaction

FO1 agent FOn agent

DSO agent

EV agent-1 EV agent-n EV agent-1 EV agent-n

Market

operator

agent

Multi Agents communication platform

JACK

A

distribution

grid

Simulink

MULTI AGENT SYSTEM

FO1

optimization slover

for grid capacity

market

participation

Matlab based

FO1

Aggregated

charging

Matlab based

FOn Aggregated

charging

Matlab based

FOn

optimization

slover for grid

capacity market

participation

Matlab based

EV user

interface

Market operator

Distribution grid

capacity market

Matlab based

Fig. 6. Multiagent system based realization of the proposed scheme of anactive power distribution network

of the market operator and the FOs. We will considersome competitive mechanisms between FOs in a futurestudy because the above case studies mainly illustratehow the FOs can cooperate to mitigate the grid congestionproblems.

According to the arguments above, we propose a multi-agent system architecture for the realization of the coordi-nation of an active distribution network with multi-actors, aspresented in Fig. 6. In which, all the agents will be built onJACK which is an agent-oriented development environmentbuilt on top of and fully integrated with the Java programminglanguage [26]. JACK offers the environment and facilitiesmessage sending/receiving. Matlab based functions enablesa declarative implementation of the decision module. Fig.7 shows the skeleton of the agents in the JACK platform,in which three agents are presented, i.e., FO agent, DSOagent, Marketoperator agent. The envelope box represents theevent that will be transferred between the agent. The roundedrectangle box means the plan that each agent has which willbe used to handle the events. Basically, this skeleton illustratesthe interactions between three agents intuitively.

VII. CONCLUSION

This paper primarily propose a framework for coordinatingthe values and operation constraints of various actors in anactive distribution network. Multilevel coordination strategiesare used in this study, i.e., price based platform is used to solvethe congestion problems between DSO and FOs and directcontrol method are adopted to control the charging of EVs byFOs. We further give an argument that multi-agent based plat-form is suitable for the demonstration of the active distributionnetwork with multi-actor setting, multilevel coordinations. A

DSOMarket

Operator

VerifyGridCongestion

NotifyCongestion

ResponseCongestion

FO_PowerSchedule

ShadowPricetoFO

MarketOperation

FO_ScheduleAD

SendPowerSchedule

FOPowerScheduleFrom EV

FO_PowerScheduleAggr

eation

posts

handles

handles

sends

handles

handles

handles

handles

handles

handles

handles

sendssends

sends

sends

sends

uses

uses

uses

uses

Fig. 7. Design view of the multi-agent systems in the JACK

scheme of the multi-agent system is presented. It is believedthat we are able to show the interactions between differentagents in an easier way by using multi-agent technology. Also,it is easier for software development. More than this, we wantto show the scalability of expanding the system into a levelwhere agents sit in different place, this could demonstrate morerealistic case.

ACKNOWLEDGMENT

The authors are grateful to the financial support of theDanish iPower project (http://www.ipower-net.dk/).

REFERENCES

[1] “Manedlig elforsyningsstatistik, html-spreadsheet summary tab b58-b72,” Danish Energy Agency, 18 January 2012. Accessed: 11 March2012.

[2] “The guardian:”denmark aims to get 50% of all electricity from windpower”, 26 march 2012,” 2012.

[3] L. E. Jones, “Strategies and decision support systems for integratingvariable energy resources in control centers for reliable grid operations,”tech. rep., Alstom Grid Inc., 2012.

[4] O. Sundstrom and C. Binding, “Flexible charging optimization forelectric vehicles considering distribution grid constraints,” IEEE Trans-actions on Smart Grid, vol. 3, no. 1, pp. 26–37, 2012.

[5] J. Lopes, F. Soares, and P. Almeida, “Integration of electric vehiclesin the electric power system,” Proceedings of the IEEE, vol. 99, no. 1,pp. 168–183, 2011.

[6] R. Hidalgo, C. Abbey, and G. Joos, “A review of active distributionnetworks enabling technologies,” in Power and Energy Society GeneralMeeting, 2010 IEEE, pp. 1–9, IEEE, 2010.

[7] Z. Xu, M. Gordon, M. Lind, and J. Ostergaard, “Towards a Danish powersystem with 50% windSmart grids activities in denmark,” in Power &Energy Society General Meeting, 2009. PES’09. IEEE, pp. 1–8, IEEE,2009.

[8] E. Larsen, D. Chandrashekhara, and J. Ostergard, “Electric vehicles forimproved operation of power systems with high wind power penetra-tion,” in Energy 2030 Conference, 2008. ENERGY 2008. IEEE, pp. 1–6,IEEE, 2009.

[9] D. Chandrashekhara, J. Horstmann, J. Østergaard, E. Larsen, C. Kern,T. Wittmann, and M. Weinhold, “Vehicle to Grid (V2G) in Denmark-Feasibility Study,” tech. rep., Technical University of Denmark DK-Kgs.Lyngby, 2008.

[10] J. Northcote-Green and R. G. Wilson, Control and automation ofelectrical power distribution systems. Crc, 2006.

[11] L. H. R. et al., “Managing congestion in distribution grids-market desingconsideration, as part of the ready project,” tech. rep., NEAS Energy,2012.

[12] N. O’ Connel, Q. Wu, J. stergaard, A. Nielsen, S. Cha, and Y. Ding,“Electric vehicle (ev) charging management with dynamic distributionsystem tariff,” in proceeding of 2011 ISGT, 2011.

[13] R. D. Zimmerman, “Uniform price auctions and optimal power flow,”tech. rep., Matpower Technical Note 1, February 2010.[Online]. Avail-able: http://www. pserc. cornell. edu/matpower/TN1-OPF-Auctions. pdfF, 2010.

[14] B. Biegel, P. Andersen, J. Stoustrup, and J. Bendtsen, “Congestionmanagement in a smart grid via shadow prices,” in Proceedings of the 8thIFAC Symposium on Power Plant and Power System Control, (Toulouse,France), Sept. 2012.

[15] D. Callaway and I. Hiskens, “Achieving controllability of electric loads,”Proceedings of the IEEE, vol. 99, no. 1, pp. 184–199, 2011.

[16] K. Heussen, S. You, B. Biegel, L. H. Hansen, and K. B. Andersen, “In-direct control for demand side management a conceptual introduction,”in procedding of 2012 Europe ISGT, Berlin, Germany., 2012.

[17] J. Hu, S. You, J. Oestergaard, M. Lind, and Q. Wu, “Optimal chargingschedule of an electric vehicle fleet,” in Universities’ Power EngineeringConference (UPEC), Proceedings of 2011 46th International, pp. 1–6,VDE, 2011.

[18] A. Mohsenian-Rad, V. Wong, J. Jatskevich, R. Schober, and A. Leon-Garcia, “Autonomous demand-side management based on game-theoretic energy consumption scheduling for the future smart grid,” IEEETransactions on Smart Grid, vol. 1, no. 3, pp. 320–331, 2010.

[19] Z. Ma, D. Callaway, and I. Hiskens, “Decentralized charging controlfor large populations of plug-in electric vehicles,” in IEEE TransactionsControl systems technology, IEEE, 2011.

[20] A. Faruqui, R. Hledik, A. Levy, and A. Madian, “Will smart pricesinduce smart charging of electirc vehicles?,” Discussion Paper, 2011.

[21] P. B. Andersen, J. Hu, and K. Heussen, “Coordination strategies fordistribution grid congestion management in a multi-actor, multi-objectivesetting,” in procedding of 2012 Europe ISGT, Berlin, Germany., 2012.

[22] S. Boyd and L. Vandenberghe, Convex optimization. Cambridge UnivPr, 2004.

[23] S. Boyd, L. Xiao, A. Mutapcic, and J. Mattingley, “Notes on decompo-sition methods,” Notes for EE364B, Stanford University, 2007.

[24] J. Hu, S. You, M. Lind, and J. Østergaard, “Coordinated charging ofelectirc vehicle for congestion preventio in the distribution grid,” IEEETransactions on Smart Grid, 2013, In submission.

[25] C. Rehtanz, Autonomous systems and intelligent agents in power systemcontrol and operation. Springer, 2003.

[26] N. Howden, R. Ronnquist, A. Hodgson, and A. Lucas, “Jack intelli-gent agents-summary of an agent infrastructure,” in 5th Internationalconference on autonomous agents, 2001.


A.8 A clearinghouse concept for distribution-level flexibility services

This paper was presented at IEEE Innovative Smart Grid Technologies (ISGT)conference, Copenhagen, Oct 2013.

A Clearinghouse Concept for Distribution-LevelFlexibility Services

Kai Heussen, Daniel Esteban Morales Bondy, Junjie Hu, Oliver GehrkeDepartment of Electrical Engineering

Technical University of DenmarkEmail: {kh, bondy, junhu, olge}@elektro.dtu.dk

Lars Henrik HansenDONG Energy

Gentofte, DenmarkEmail: [email protected]

Abstract—Flexibility resources on the demand side are an-ticipated to become a valuable asset for balancing renewableenergy fluctuation as well as for reducing investment needs indistribution grids. To harvest this flexibility for distribution grids,flexibility services need to be defined that can be integrated withdistribution grid operation and that provide a benefit that can betraded off against other grid investments. Two key challenges arehere that the identification of useful services is still ongoing andthat the transaction cost for the individually small contributionsfrom the demand side could be prohibitive. This paper introducesa flexibility clearinghouse (FLECH) concept and isolates FLECHkey functionality: to facilitate flexibility services in distributiongrids by streamlining the relevant business interactions whilekeeping technical specifications open.

I. INTRODUCTION

Procuring system services from small distributed energyresources (DER), including flexible demand, has become afocus of research and field trials in recent years. The capabilityof managing DER via aggregators, to participate in bulk powermarkets as well as to provide system (ancillary) services hasachieved much progress. Yet the provision of such services atdistribution level has not seen the same development, either be-cause they are simply uneconomic or possibly because market-based approaches are not easily introduced to distributionsystem operations.

A. Demand Response and Distribution Grid Planning andOperation

Two new drivers for increased grid demand are: 1) thedevelopment of renewables, in particular solar photovoltaicinstallations, and 2) the introduction of controllable and likelyprice responsive power consuming/producing units, includingheat pumps and electric vehicles. These new units may causegrid issues such as voltage violations, reverse power flows,or thermal overloading. Such grid issues force distributionsystem operators (DSOs) to consider costly reinforcementswhile the effective grid utilization may actually be decreasingdue to reduced diversity amongst the price responsive loads.Distribution grid planning and operation are conventionallyfocussed on planning and maintenance and are only slowlyadopting “smart” approaches involving automation. The smartgrid is suggested to enable better infrastructure utilization andthe accommodation of additional generation and demand. Onemodern approach to increasing the effective hosting capacity

of distribution grids is to improve situational awareness forgrid operators and planners through better feedback about theoperating state [1].

Another approach is a better utilisation of demand sideflexibility through new demand response programs, or ‘flex-ibility services’ [2]. Aggregating and controlling DER forcommercial purposes is a field for new actors where techno-logical and business innovation are essential for development.This development will either be facilitated or hampered byregulatory choices. The need for a regulatory framework forcommercial flexibility services at the distribution level hasbeen recognized, e.g. by the standardization mandate M490and with the introduction of the traffic light concept [3], andis under development.

The introduction of new flexibility services at distributionlevel is thus a current and relevant concern, while, clearly,both the service definitions and the potential market-basedcoordination of such services are still largely open problems.

B. Congestion Management in Distribution Grids

At transmission level, congestion constraints are a commontool for reflecting transmission limitations toward marketmechanisms, e.g. when interconnecting market areas. Conges-tion occurs when scheduled energy flows exceed the availabletransmission capacity [4]; congestion management is thenthe allocation of transfer capacity according to economicprinciples. Based on this definition, congestion can also bedefined for distribution networks [5] as a coordination strategyreflecting operational constraints on market terms. A study in[5] analyzed three kinds of market mechanisms for alleviatingcongestion. In [6]–[8], prices are used to coordinate betweenDSO, aggregators and DER owners to achieve an optimalallocation, and in [9] different grid tariffs are discussed.

A simplification common in the distribution congestionliterature [5]–[9] is to focus on ‘bottleneck’ constraints on asummation of power flows. This constraint can be interpretedas transformer current limit - for which there exists a businesscase on deferring grid investments. However, a DSO’s actualdecision drivers and investment alternatives are often morevaried and complex than the ones considered here, and regu-latory requirements often inhibit the application of congestionconstraints in distribution grids.

1

To achieve an incremental adoption and a better matchof flexibility services with DSO regulations and procedures,DSOs should be enabled to request flexibility services adaptedto actual practices of distribution system planners and to needsof flexibility service providers.

This paper introduces a concept which respects the practicalrequirements of a technically advancing DSO while openingthe efficiency potential of market-based congestion manage-ment: the Flexibility Clearinghouse (FLECH). Following threekey design principles, 1) to minimize transaction cost forDSO flexibility services; 2) to allow for further technicalspecifications of DSO services; and 3) to focus on businesstransactions and do not interfere with with distribution op-erations, this concept has been developed and implementedwithin the Danish research and innovation project iPower:www.iPower-net.dk.

II. FLEXIBILITY SERVICES FOR DISTRIBUTIONOPERATION

This Section defines the concept of a Flexibility Service,provides examples, and a framework for analysis.

A. Definition of a Flexiblity ServiceMany DER units have the capability of altering their

generation/consumption pattern with limited impact on theirprimary energy service. This capability is further referredto as DER flexibility. Flexibility can be provided to a DSOthrough a new dedicated ancillary services market to whichentities representing DER, here Aggregators, can submit bids.The products on this market are called flexibility services andinclude a detailed specification of the service procurement,activation, delivery, validation and settlement. These servicesinclude two generic types: (A) fully scheduled services whichoblige the aggregator to behave as contracted without DSOintervention, and as well as (B) reserve services which entaila reserve or availability combined with a need-based activationby the DSO.

B. DSO Flexibility Service ExamplesAn analysis of relevant issues in the distribution system,

reported in [2], identifies four key needs that could be fulfilledby flexibility services: response to foreseen and unexpectedoverloading, fast response to resolve N-1 situations, support incase of voltage limit violations (power quality), and supportwith respect to reactive power exchange with the transmissiongrid.

The same report, [2], suggests seven potential flexibilityservices to support the above needs. This paper will focuson those five services which offer flexibility via active powermanagement:

1) PowerCut Planned: Used for handling predictable peakload for periodically daily issues in advance.

2) PowerCut Urgent: Used for handling peak loads on anevent basis.

3) Power Reserve: Used when the system is operating inthe reserve band of the feeder, and a fault in the systemwould require the utilization of the reserve band.

4) PowerCap: Activated upon request to ensure that thecapacity limits specified by the DSO are not violated.

5) PowerMax: Same function as PowerCap, but activatedthrough a planned schedule.

These services address the first two needs mentioned above,i.e. response to overloading and to N-1 situations. Notably,they include both fully scheduled products, i.e. (1) and (5),as well as reserve products, (2)-(4). As stipulated in [2],these service definitions are expected to be among the firstones accepted by DSOs. However, they do not constitute anexhaustive list of potential services.

C. A Framework for Analyzing Flexibility Services

In [5] an analysis framework of four stages for has beenintroduced: 1) Offline Planning, 2) Online Scheduling, 3) Real-time Operation, 4) Offline Settlement.

The framework is suited to identify alignement of technicaland market functions across all participating actors. Keyoperations for each stage are listed in Table I.

TABLE ISTAGES OF SERVICE PROVISION AND ASSOCIATED FUNCTIONS

Stage Market function Technical function

Offline Planning Contract specificationand cost allocation

Grid planning and ser-vice specification

Online Scheduling Contracting and re-source allocation

Scheduling and reser-vation

Real-time Operation Contractual fulfillment Plan execution and ac-tivation/response

Offline Settlement Financial settlement Service validation

The stages and the separation of market and technical oper-ations define a framework suited for the analysis of flexibilityservices and isolation of FLECH functionality.

III. THE FLEXIBILITY CLEARINGHOUSE CONCEPT

As described in [2], the FLECH is meant to facilitate DSOsto announce services and aggregators to bid upon. Here, thestakeholder setting and FLECH core functions are identified.

A. Stakeholder Roles and Need for a Flexiblity Clearinghouse

Demand for system services from DER units exist all theway down to the low voltage grid. We identify associated in-terests with respect to the following conventional stakeholders:

• Transmission System Operator (TSO)• Distribution System Operators (DSOs)• Balance Responsible Parties (BRP)

New stakeholders in the context of DER services include:• DER owners• AggregatorsAll stakeholders have interests of their own that require

alignment to enable successful delivery of flexibility of flex-ibility services to a DSO. The DER owner is interested inoffering flexibility which does not negatively influence theprimary function of the unit. This flexibility will be definedin contracts between the DER owner and an Aggregator.

2

Flexible Demand (DER)

Electricity Spot Market

Grid Measurements

Fleet Operator


control

Information flow

Control relation

Conventional Demand

Ancillary Services

Physical connection

Aggregation

Smart Meter Data

DER Owner

Distribution system

operator

Transmission System

OperatorOther market participants

Aggregator

FLECH

FLECH operator

Electricity Spot Market

AncillaryServices

TransmissionSystem

OperatorpOther market participants

Fig. 1. Schematic overview of the considered actors and their roles in relationto the FLECH.

Depending on the capabilities of a particular DER unit, it maybe able to offer system services to more than one interestedparty. For example, a controlled decrease in consumption couldeither be part of a frequency control service offered to a TSO,or it may be offered to a DSO for peak shaving. In any case,the invocation of the service would impact both grid domains,and the interests of TSO and DSO could conflict. In the presentelectricity market regulation, larger DER units interact witha TSO through a BRP. The business model of a BRP addsanother set of interests which do not automatically align withthe requirements of grid operation.

Two present ways of addressing flexiblity services are gridcodes and bilateral contracts. Grid codes are primarily suitedto define absolute limits of operation. They have to be rigidas their scope is universal to all grid connected devices.The creation and updating of grid codes is a slow processwhich is not very innovation friendly and does not imply anyremuneration. Individually negotiated bilateral contracts, onthe other hand, imply a high transaction cost, which makesthem unsuitable for services with small economic margins.

This setting illustrates the need for a service-independentand open platform for the arbitration of interests between thestakeholders involved in delivering a power system servicefrom flexible DER units. The scope of this platform shouldbe limited to providing an infrastructure on top of which thebusiness logic of present and future services can be defined.

B. Functions of FLECH

A clearinghouse is a safeguard for a marketplace, ensuringthe secure fulfilment of a business transaction. FLECH pro-vides the platform for such transactions with respect to smart

services in the interest of a distribution system operator.The FLECH concept is realized as a service-oriented plat-

form that facilitates the business process of specifying, con-tracting, delivery and settlement of DER flexibility services.It requires involvement of a software provider. For operation,a new neutral stakeholder, similar to the role of a marketoperator for the bulk electricity markets, should be introduced.

The capability of FLECH will evolve with the develop-ment of distribution level markets. Initially, market clear-ing can be performed by the DSO. The FLECH function-ality would mainly consist of bookkeeping and communi-cation/broadcasting functions. In the future, when marketmechanisms are more stable and services well-defined, marketclearing would be implemented on the FLECH platform.Another optional functionality is a coordination role in thescheduling phase. None of the specified services in [2] requiresuch functionality as it is allocated to DSO and Aggregatorsinternally. For future congestion management strategies, suchas reported in [5], online coordination of several aggregatorswould theoretically be more economically efficient.

During operation, two alternative service types – reserveand scheduled – need to be distinguished. To avoid technicalreal-time requirements for FLECH, the activation of reserveservices should be sent directly to the respective aggregator,while FLECH assumes a pure bookkeeping role.

Finally, FLECH supports service validation and settlement.As all records of activations are available, FLECH can matchbids and fulfillment and calculate the final settlement.

IV. FLECH INTERACTIONS

The FLECH functionality outlined above aims to facilitateinteractions between DSO and flexiblity service providers; thissection identifies the required interactions and isolates thecommon message exchange requirements.

A. Flexibility Service Mapping

The framework introduced in Section II-C is used tomap out the FLECH interactions for the candidate flexiblityservices summarized in Section II-B. Here, two services,PowerCut Urgent and PowerMax, are chosen as representativecases and their mapping is summarized in Tables II and III.

TABLE IIPOWERCUT URGENT


Offline Planning Specification and an-nouncement of reservecontract

DSO: identify locationand volume of reserveneed

Online Scheduling (optional: call for shortterm bids to activationmarket)

DSO: Anticipate ’ur-gent’ activation needtime

Real-time Operation – DSO: Activationsignal; DER respondwithin 15min

Offline Settlement Payment per activa-tion. Failure to de-liver 4 times termi-nates contract.

Recording and valida-tion of activation sig-nal and response

3

DSO FLECH Aggre-gator

Planning

Settlement

submit tenderservice X, area Y announce

tender area Y

register area Y

confirm

submit bid ID Zbid list area Y

accept list area Yaccept bid ID Z

Operation

scheduled service

reserve service

Scheduling

optional loop

perform service

notification ID Znotification

ID Z

activation ID Z perform serviceregister

activation ID Z register performance

ID Zconfirm confirm

metering data*

settle ID Z

confirm ID Zrequest settlement

validation

confirm validation ID Z

forward settlement ID Z

confirmconfirm

cleared bids, a. Y

Fig. 2. Sequence diagram for FLECH communication

TABLE IIIPOWERMAX


Offline Planning Announce service re-quest

Identify activationlocation, periods andvolume

Online Scheduling (optional coordination) (DER and Aggregator:prediction)

Real-time Operation – DSO: monitoringonly; Aggregatorcoordinates DER

Offline Settlement Payment defined in thecontract. Failure to de-liver 4 times termi-nates contract

Recording and valida-tion of activation sig-nal and response

The following conclusions can be drawn from the analysis:At the planning stage, tender announcement and reserveagreements with a long lead-time are arranged, e.g. 2 - 6months before activation. The scheduling phase is optional, yetvaluable in a competitive setting: due to short lead time (e.g1 - 24h ahead) bid prices would improve due to reduced uncer-tainty. At the operation stage, the services can be differentiatedinto reserve and scheduled services. For scheduled services asimple notification of activation will be performed. For reserve

services, the DSO will send an activation signal directly tothe aggregator. FLECH must be notified if the activationis executed. The notifications are used for settlement. Withregards to offline settlement, FLECH will be the responsiblefor coordinating validation, consolidating the judgment fromdifferent actors.

B. Generic FLECH Messages

The FLECH key interfaces are to DSO and Aggregator.The sequence diagram in Figure 2 summarizes the essentialmessage flow, grouped by stage. Focussing on the transac-tional, administrative, aspects of service provision, the keyinteractions are common to all services considered, with twoexeptions: the scheduling stage would depend on the respectiveservice and market clearing model, and in the operation stage,separate sequences are defined for scheduled and reserveservices. The adoption of new services to the a first FLECHdesign will therefore come at a small incremental cost.

V. CASE STUDY

This case study illustrates the application of the PowerMaxservice by a DSO in case of an anticipated low voltagetransformer overload. Consider the following scenario:

(A) 70 household consumers are connected to a 10/0.4kVtransformer T1, each with connection capacity of approxi-mately 7kW ; an electric vehicle (EV) with up to 3.7kW charg-ing capacity, [10], is associated with 14 of the households.

(B) Aggregators managing controllable consumption in thisgrid area are two EV fleet operators, FO1 and FO2. FO1operates 5 EVs and FO2 9 EVs, corresponding to 18.5kWand 33.3kW charging capacity, respectively.

(C) Based on historical data and specific load models, theDSO anticipates that the 175kW limit (corresponding to 70%of the maximum 250kW ) of transformer T1, may be exceededby 37.8kW on weekdays between 4:30pm and 8:00pm duringthe months of December, January and February, mainly causedby additional EV charging loads; the corresponding loadprofile is illustrated in Figure 3.

(D) The DSO has the option to reinforce the transformeror to acquire a flexibility service. An economical evaluationsuggests that flexibility services could postpone reinforcementand thus are an attractive option. Due to the characteristics ofEV loads, the PowerMax service is chosen as most viable.

To prepare the FLECH tender, the DSO identifies its needs:having 14 EVs with flexible consumption in the area chargingat maximum rate, the peak load of the EVs is 51.8kW –under current Danish regulations, the only capacity limit is thephysical connection capacity. The DSO thus needs to reduce,for the given time window, this maximum capacity limit to14kW , i.e. the total capacity of the two aggregators shouldbe reduced by 37.8kW . With PowerMax, the DSO thereforerequests a reduction against documented connection capacityof the aggregators in the area. Apart from announcing thequantity to be reduced, the DSO includes a ‘recommended’rate in order to initiate price discovery.

4

0 5 10 15 20 250

50

100

150

200

Time/h

Pow

er/k

W

0 5 10 15 20 250

50

100

150

200

Time/h

Pow

er/k

W

FO1 load

FO1+FO2 loadBase load

Transformer loadControllable cap.

Transformer cap.

14 kW

Fig. 3. Expected load profiles for days with anticipated feeder overload.Top: Expected profile before PowerMax tender. Bottom: Expected profile afterPowerMax tender.

In this scenario, FLECH facilitates the tender and trans-actions associated with the service provision. The individualsteps are related to Figure 2. At the planning stage, the DSOsubmits the following service tender to FLECH:PowerMax:

CAPACITY REDUCTION [AREA T1]: 37.8kW

TIME 4:30pm TO 8:00pm ON weekdays

PERIOD 01 Dec 2014 TO 28 Feb 2015

RECOMMENDED RATE 500 EUR/kW

This tender is then announced by FLECH to all aggregatorsregistered for T1. The aggregators bid into the FLECH:AggID [BidID]: reduction FROM capacity AT rate flex?



FO2[FO2B2]: 15.4kW FROM 18.5kW 1000 EUR/kW FLEX

Note that FO1 did not bid with all of its resources, effectivelyonly using 4 out of 5 cars, and that the second bid by FO2is FLEX bid, i.e. it does not need to be accepted entirely.After gate closure their bids are forwarded to the DSO whichevaluates the offers and decides to accept the following bids:BidID: FO1B1, FO2B1, FO2B2*90% AT 1000 EUR/kW

This leads to an effective capacity reduction of 38.5kWwhich fulfills the required 37.8kW . The prices of this casestudy are completely fictitious and not anchored in real costs.

As the PowerMax service includes a schedule, aggregatorsdetermine their commitment internally and no interaction isrequired at the online scheduling stage. At the operation stageFLECH collects the activation notifications which are passedon to the DSO to verify the performance. At the settlementstage, the Metering Responsible submits metering data toFLECH and FLECH validates the performance. The settlementtransaction is facilitated by FLECH.

VI. CONCLUSION

This paper presented a clearinghouse concept for facilitatingancillary services at the DSO level. With the emergence ofnew players in DSO ancillary service markets, it is foreseenthat such a mechanism will be needed to minimize transactioncosts. In contrast to other contributions on distribution conges-tion mitigation, the FLECH adapts to the actual DSO needsand is not tied to a specific aggregator architecture

Two representative flexibility services have been chosento identify the FLECH requirements. By separating marketand technical aspects of the services, it is possible to isolatethe need for a pure market facilitator, which either facilitatesbilateral contracts or operates a market-clearing facility.

The role of FLECH and its interactions with stakeholders ofthe distribution flexibility service market have been described.A case study has been presented showing how FLECH isenvisioned to work in a concrete scenario.

Several important aspects of flexiblity services have beenout of the scope of this paper and will be addressed in futurework: (A) A flexiblity service requires the formulation of a‘baseline’. This baseline at transmission level is based on theenergy markets, but there is no such formal baseline at thedistribution level for DER. (B) The activation of flexibilityservices may potentially cause an imbalance cost. It is not clearto which actor this cost should be allocated. (C) The aggregatoris assumed to be an independent entitiy representing the DERtowards FLECH and the DSO. This assumption is currentlynot backed up by regulation, partly because of the imbalanceissue noted in point (B).

VII. REFERENCES

[1] P. VINTER and H. KNUDSEN, “Using continuous state estimation ingrid planning,” in In Proceedings of the 20th International Conferenceon Electricity Distribution Pragu, 8-11 June 2009.

[2] N. C. Nordentoft, Y. Ding, L. H. Hansen, P. D. Cajar, P. Brath, H. W.Bindner, and C. Zhang, “Development of a dso-market on flexibility ser-vices,” iPower, http://ipower-net.dk/Publications.aspx, Tech. Rep., 2013.

[3] Smart grid use case management process. CEN-CENELEC-ETSI SmartGrid Coordination Group - Sustainable Processes.

[4] M. Shahidehpour and Y. Wang, Communication and control in electricpower systems: applications of parallel and distributed processing.Wiley-IEEE Press, 2004.

[5] P. Bach Andersen, J. Hu, and K. Heussen, “Coordination strategies fordistribution grid congestion management in a multi-actor, multi-objectivesetting,” in ISGT Europe (Berlin). IEEE, 2012, pp. 1–8.

[6] B. Biegel, P. Andersen, J. Stoustrup, and J. Bendtsen, “Congestionmanagement in a smart grid via shadow prices,” in Proc. 8th IFACSymp. on Power Plant and Power System Control, Toulouse, 2012.

[7] E. L. Karfopoulos and N. D. Hatziargyriou, “A multi-agent system forcontrolled charging of a large population of electric vehicles,” IEEETransactions on Power Systems, 2012.

[8] J. Greunsven, E. Veldman, P. Nguyen, J. Slootweg, and I. Kamphuis,“Capacity management within a multi-agent market-based active distri-bution network,” in Innovative Smart Grid Technologies (ISGT Europe),2012 3rd IEEE PES International Conference and Exhibition on. IEEE,2012, pp. 1–8.

[9] L. H. Rasmussen, C. Bang, and M. Togeby, “Managing congestion indistribution grids-market design consideration,” Ea Energy Analyses,Tech. Rep., 2012.

[10] F. Marra, G. Y. Yang, C. Traholt, E. Larsen, C. Rasmussen, andS. You, “Demand profile study of battery electric vehicle under differentcharging options,” in PES General Meeting, 2012 IEEE, 2012, pp. 1–7.

5


A.9 A multi-agent system for distribution gridcongestion management with electric vehi-cles

This paper is to be published in Journal Engineering application of artificialintelligence, Volume 38, Feb 2015, Page 45-58.

A multi-agent system for distribution grid congestion management with electric vehicles

Junjie Hua,∗, Arshad Saleemb, Shi Youa, Lars Nordstromb, Morten Linda, Jacob Østergaarda

a Center for Electric Power and Energy, Department of Electrical Engineering, Technical University of Denmark, Denmark.b Department of Electrical Engineering, Royal Institute of Technology, Stockholm, Sweden.

Abstract

Distributed energy resources, like electric vehicles (EVs) are widely regarded as valuable assets in the smart grid in addition to theirprimary transport function. However, connecting EVs to the distribution network and recharging the EV battery without any controlmay overload the transformers and cables in peak hours when EVs’s penetration is relatively high. In this study, a distributed controlstrategy for integrating EVs into the distribution network is proposed to coordinate the self interests and operational constraintsof two actors: the EV owner and the Distribution system operator (DSO), facilitated by the introduction of fleet operator (FO)and grid capacity market operator (CMO). The control strategy includes control system design and distributed control algorithmdevelopment which is based on general equilibrium market mechanisms. In order to fully demonstrate the coordination behaviorinside the proposed strategy, we build a multi-agent system (MAS) which is based on the co-simulation environment of JACK andMatlab. A use case of the MAS and the results of running the system are presented to intuitively illustrate the effectiveness of theproposed solutions.

Keywords: Congestion management, Distribution grid, Electric vehicles, Multi-agent system, Resource allocation.

1. Introduction

EV is widely advocated as a mean of personal transport andurban delivery, since it can contribute to the reduction of CO2emission, especially when the recharging electricity is gener-ated by renewable resources. However, for the electric utilities,the issue is how to integrate the EVs smoothly into the grid, i.e.,manage the simultaneous charging of a large scale of EVs, with-out overloading the grid. Several studies (Heydt, 1983; Lopeset al., 2011; Clement-Nyns et al., 2010; Green II et al., 2011)have shown that uncontrolled charging (alternatively is calleddumb charging) of EVs will challenge the capacity of the dis-tribution grid. To address this challenges, the time-of-use tariffsor multiple tariffs charging scheme are used in the early stageto relieve the congestions in the peak hours (Shao et al., 2010).But using tariffs solely are not adequate to eliminate the con-gestion, since they would shift peak load to its neighbouringperiod (Ma et al., 2013; Karfopoulos and Hatziargyriou, 2013).

Recently, much research has been aiming to coordinate theobjectives and the constraints centrally, e.g., to optimize thecharging cost of EVs as well as respecting the hard constraintsimposed by EV owner needs and distribution grid operation. In(Sundstrom and Binding, 2012), a complex scheduling probleminvolving the EV owners, the fleet operator (FO) and the DSO isanalyzed. The approach requires a complex interaction betweenthe DSO and the FO, on each interaction, the FO will get a spe-

∗Corresponding author. Tel:+45 45253487Email addresses: [email protected] (Junjie Hu),

[email protected] (Arshad Saleem), [email protected] (ShiYou), [email protected] (Lars Nordstrom), [email protected](Morten Lind), [email protected] (Jacob Østergaard)

cific grid constraint from the DSO and add it into the EV charg-ing cost minimization problem. The results show that both theFO and the EV owners can achieve the objectives of minimiz-ing charging costs and fulfilling driving requirements withoutviolating the grid constraints. Lopes et al. (Lopes et al., 2009)proposed a conceptual framework consisting of both a technicalgrid operation strategy and a market environment to integrateEVs into the distribution systems. The FO is proposed to man-age the EVs and the FOs will prepare the buy/sell bids into theelectricity market. Having this defined, a prior interaction withthe DSO must exist to prevent the occurrence of congestion andvoltage problem in the distribution network. The smart charg-ing algorithm is mainly designed for the operation of the DSOwhich can maximize the density of the EV deployment into thegrid. It is also assumed that the grid has enough capacity toprovide all the power required by EVs. With this assumption,the centrally smart charging approach is effective.

Although these proposed solutions are shown to work effi-ciently for a limited number of EVs, centralized managementrequires the acquisition and processing of an enormous of in-formation for a large penetration of EVs, such as: 1) the batterymodel of EVs, initial state of charge (SOC) and desired SOCof EVs; 2) driving pattern of EVs; 3) grid constraint informa-tion from DSO; 4) electricity market information, which wouldsubstantially request significant computational resources, com-munication overheads and communication infrastructure cost.Research in (Lyon et al., 2012) indicates that the benefits for thecentralized charging management might not be justified for thecommunication infrastructure cost. Alternatively, several waysof solving the congestion problem in the distribution grid havebeen suggested from market perspective. In (Andersen et al.,

Preprint submitted to Engineering Application of Artificial Intelligence December 5, 2013

2012), the paper conceptualizes several approaches, i.e., dis-tribution grid capacity market, dynamic grid tariff (O’Connellet al., 2011), to address the distribution grid congestion. Theconceptualized strategies for congestion management are eval-uated in terms of their complexity of implementation, the valueand benefits they can offer as well as possible drawbacks andrisks. Further on, the work in (Hu et al., 2013) analyzed theshadow price based grid capacity market scheme, in which, FOscentrally schedule and control the charging of EVs, they negoti-ate with the market operator (distribution grid capacity market)on the limited capacity of the distribution grid if it is needed.The focus of the study (Hu et al., 2013) is the mathematicalproof of the proposed market scheme. Some numerical casestudies are shown to illustrate the effectiveness of the proposedsolution.

To implement and assess both control strategies of smartcharging of EVs, especially market based coordination method,a multi-agent system (MAS) based technology is very suitable(Jennings and Bussmann, 2003), which can be justified by thefollowing reasons:

• The increase in complexity and size of the whole EVcharging network bring up the need for distributed intelli-gence and local solution, which fall into the scope of MASbased technology.

• The information flow, optimizations and the negotiationshappened in the smart charging network of EVs can bewell demonstrated and integrated into a MAS.

• The system can be pre-tested and pre-analyzed by using aMAS before going to real implementation.

In addition to these general arguments, MAS has been widelyproposed in the context of power systems, such as power sys-tem restoration (Nagata and Sasaki, 2002), power system op-eration and control (Rehtanz, 2003). More recently, the multi-agent concept is proposed for distribution system operation andcontrol (Nordman and Lehtonen, 2005; Issicaba et al.; Pipat-tanasomporn et al., 2009; Ren et al., 2013), especially, con-sidering the capacity management with a large population ofelectric vehicles (Karfopoulos and Hatziargyriou, 2013; Mi-randa et al., 2011) and the capacity management with more gen-eral loads (Greunsven et al., 2012). The authors (Karfopoulosand Hatziargyriou, 2013) proposed a distributed, multi-agentEV charging control method based on Nash certainty equiv-alence principle that considers distribution network impacts.Four types of agents are included in the study, EV aggregatoragent, regional aggregation agent, microgrid aggregation agentand cluster of vehicles controller agent, and vehicle controlleragents. In the non-cooperative, dynamic game, all the vehiclescontroller agents decide the strategy that minimizes his own ob-jective functions. The up-level agents coordinate vehicles con-troller agents’ charging behaviour by altering the price signal.The price signal is a reflection of congestion conditions. The re-sults indicated that the proposed approach allocates EV energyrequirements efficiently during off-peak hours which achieveseffectively valley filling and also leads to maximization of load

factor and minimization of energy losses. The authors in (Mi-randa et al., 2011) used the MAS to design a distributed, modu-lar, coordinated and collaborative intelligent charging networkwith the objective of pro-actively scheduling the charging ofup to fifty EVs as well as eliminating the grid overloading is-sue. The study mainly considered how the electricity is dis-tributed to the multiple charging point agent under one localpower manager agent and this is done by an auction mecha-nism. Each charging point agent makes a bid for the energy inthe next 15 minutes until it get the desired state of charge ofthe battery, then the local power manager agent sorts out theorders to determine which EV can be charged during the timeslot. In (Greunsven et al., 2012), an active distribution network(ADN) is presented with its actors and their objectives. Themulti-agent technology is proposed for the normal operation ofthe ADN, in which the auctioneer agent (placed at the MV/LVtransformer) communicates with the device agent by sendingthe price signal and receiving the bid curve. Further on, capac-ity management is investigated by transforming the bid curvesof the device agents.

This paper is an extension of the concepts described in (Huet al., 2013). The extension mainly utilize the MAS technologyto assess the proposed market scheme with the purpose of illus-trating and tracking the coordination behaviors. Moreover, theextension considers demonstrating EVs’ flexibility (through thepresence of response weighting factor to the shadow price) inthe developed MAS system. There are several highlights in thisstudy compared to previous studies (Karfopoulos and Hatziar-gyriou, 2013; Miranda et al., 2011; Greunsven et al., 2012).

• The developed MAS offers a modeling environment thatenables study of important characteristics of the proposeddistribution grid capacity market which is not presentlyavailable. By implementing and assessing the marketbased strategy, it is shown that grid congestion problemcan be eliminated in few steps.

• The developed MAS explicitly presents the relevantagents, the plan, and the event inside a market frame. Themodelling approach can be an example for other similarproblems.

• The developed MAS demonstrates a simulation platformwhich is based on the integration of JACK and Matlab.The platform can well integrate the advanced optimizationand control and the interactions which can vary from sim-ple information passing to rich social interactions such ascoordination, negotiation.

The remainder of the paper is organized as follows: In sec-tion II, an introduction is given on the assumptions and controlsystem architecture. Section III mainly presents the mathemat-ical principles behind the methods of smart charging of EVsand distribution grid congestion management. In section IV,MAS based realization of congestion management scheme ispresented. Case studies are illustrated in section V to facili-tate the understanding. Finally, discussion and conclusions arepresented in section VI.

2

2. Control system architecture

2.1. Main actors in the control system

Typically, the challenges in the distribution grid causedby the increasing electricity consumption from EVs and heatpumps (Søndergren, 2011) are solved by expanding the grid tofit the size and the pattern of demand. As an alternative, in-spired by the congestion management method at the transmis-sion system level, it is then defined in this study that the capac-ity of the distribution network (scared resources) is allocatedaccording to economic principles without upgrading the grid.

Fig.1.(a) presents a sketch of a typical situation in a distribu-tion network where the substation supports the electricity to thehouseholds connected to it. In this distribution network, it isassumed that the consumers own controllable appliances, i.e.,EVs, besides some conventional loads. These EVs have con-tracts with the FOs who are the new entities in a smart grid en-vironment. FO has been widely proposed to provide the chargeservices to EVs and it is further assumed that the FO is alsoresponsible for managing the EV charging infrastructures, i.e.,the EV supply equipment (EVSE) (Bessa and Matos, 2012; SanRoman et al., 2011). As illustrated in Fig.1.(b), EVSE supportsthe smart charging functions. The decision can be made on theEV level or on the FOs level. The IEC 15118 is the most recom-mended communication standard and demonstrated in details in(Kabisch et al., 2010; Schmutzler and Wietfeld, 2010) by show-ing the sequence diagram of a charging pro-cess between theEVSE and the EVs. For the communication between the EVSEand the FOs, it is recommended that IEC 61850 can fulfil thefunctions. We use EVi as an agent to represent the EV owner’soperation on EVs and it will communicate with the FOs. Inthis study, it is assumed that the DSO will coordinate with theFOs to alter the EV’s charging profile to prevent/eliminate theoverloading problem. The coordination between the DSO andthe FOs is facilitated by the grid capacity market operator. Inthe following section, some market based coordination methodswill be discussed for the interaction between the DSO and theFOs.

2.2. Coordination relationships between the actors in the con-trol system

2.2.1. Allocating the available power of DSO among the FOsby standard price-oriented market protocols

As discussed in (Akkermans et al., 2004; Wellman, 1993;Cheng and Wellman, 1998), market-based control method isvery efficient and applicable for handling the resource alloca-tion problem. The authors (Akkermans et al., 2004) discussedthe theoretical foundations of distributed large-scale controlproblem by unifying the microeconomics and control engineer-ing in an agent-based framework. One of the main results ofthis study is that computational economies with dynamic pric-ing mechanisms are able to handle scarce resources for con-trol adaptively in ways that are optimal locally as well as glob-ally. It is further recommended in the study (Akkermans et al.,2004) that standard price-oriented market protocol, e.g., Well-man’s WALRAs algorithm (Wellman, 1993; Cheng and Well-

Grid Measurements

DSODistribution Operation

Conventional load

Information flow

Physical connection

FOs

Public networkEV

owner

EVSE

E.g., IEC 61850E.g., IEC 15118

FOj

EVi

EVnEViEV1 EVk

FOi

(a) Main actors in the control systems

(b) Relevant ICT standard between FOs and EVs

Distribution system

operator

Grid capacity market operation

Market operator

Figure 1: A illustration of distribution network with EVs.

man, 1998) is suitable for implementing the agent-based mi-croeconomic control. The algorithm presumes an auctioneeragent which announce the market clearing price p and the con-trol agents will submit their demand γα based on the price,then the auctioneer agent updates the price until the equilibriumvalue is found. The market-based approach has bee supportedto be used in the power distribution system, such as the discus-sion in joint research center European Forum 1 or in the research(Nordentoft, 2013; Schlosser, 2010; Lorenz et al., 2009).

2.2.2. Coordination method between FOs and EV owners

The control method between the FO and the EVs developedin (Sundstrom and Binding, 2012; Lopes et al., 2009; Hu et al.,2013) belongs to centralized control strategy, while the one de-veloped in (Ma et al., 2013; Karfopoulos and Hatziargyriou,2013) goes for distributed control. Studies in (Karfopoulosand Hatziargyriou, 2013; Richardson et al., 2012) compared thecentralized control and distributed control method when utiliz-ing them to make an optimal plan which can optimally deliveryenergy to EVs as well as avoiding grid congestions. They out-lined the advantages and disadvantages of both strategies.

1European commission Joint research centre, Scientific support to capacitymarkets and the integration of renewables, Brussels (BE) - 22/07/13

3

2.3. Distribution grid congestion management with EV inte-gration

Shadow price protocols are proposed for the coordination be-tween the DSO and the FOs agent in (Hu et al., 2013), in whichthe shadow price used as a market clearing price is updated ineach bidding round by the grid capacity market operator (servedfor the DSO). The bids is coming from FOs which representthe EVs and directly schedule and control the charging of EVs.In this study, we modify the coordination method between theFOs and the EVs by distributing the charging decision to theEV agent. A response weighting factor to the shadow price isintroduced to the individual EV agent. With this presence, theEV agent can show their willingness to charge or not during thehigher price time slot.

In the following section, a distributed method is proposed forintegrating the EVs into the power distribution systems. Fig. 2shows the steps of the proposed methods:

1. The EV owner selects charging requirements and EVcontroller generate the charging schedule, e.g., based oncharging least cost strategy, dumb charging strategy etc.

2. The EV owner sends the charging schedule to the FOwhich they have been subscribed.

3. The FOs aggregate the charging schedule from their con-tracted EV owners and submit the aggregated chargingschedule to the DSO.

4. The DSO verifies the charging schedule of FOs by runningload flow calculation and sends the results to all the FOs.

5. The FOs submits the charging schedule to the Market op-erator if there exists congestions; otherwise, the FOs couldbid the energy schedule to the energy spot market and theprocedure stops in this step.

6. The Market operator sends the shadow price to the FOsand the FOs resubmits the charging schedule to the Marketoperator until the shadow price is converged.

7. The FOs sends the shadow price to all the EV controllers.8. Repeat the step 1 to step 7 until the congestion is totally

eliminated in the planning period.9. Bid final energy/power schedule to the electricity spot

market.

The key concept is that the FOs/EVs’ energy schedules are co-ordinated by the DSO/Market operator before they are sent tothe energy market.

2.4. Further discussion on the proposed methodWith the purpose of further illustrating the proposed distri-

bution grid capacity market, we give some basic introduction tothe congestion management method and the markets, i.e., Spotmarket and Regulating power market, operated at the transmis-sion system level. Three very different methods of managingthe congestion of the transmission system in the deregulated en-vironment, have been presented in (Christie et al., 2000). Thethree methods are the optimal power flow model used in theUnited Kingdom, Australia, New Zealand, and some part of theUnited States, the price area based model used in the Nord-Pool market area in Nordic countries and the transaction based

1. EV driver selects charging requirements and EV

controller make the charging schedule

FOs

2. Charging schedule request from EV to FO

Public network

Distribution system

operator

Market operator

7. FO sends the shadow price to EV controller

3. FO submits the aggregated charge schedule to DSO

4. DSO verifies charging schedule of Fos.

5. DO submits the charge schedule to market operator

6. Market operator sends the shadow price to Fo

EV owner

Figure 2: Scheme of the proposed solutions

model used in the United states. In Spot market, the Power Bal-ance Responsible parties (PBRs) make the power and energybids into the market, consists of conventional power and windpower. With the trading, they can balance the power systemsin the deregulated environment. As electricity production andconsumption always have to be in equilibrium, deviations inthe operating hours are left for the transmission system oper-ator (TSO) to balance which is done via the regulating powermarket. Note that the dispatch currently is set only based on thespot market and the operational state of the distribution grid isnot considered. In short, the proposed solution of this study canenable the distribution congestion management before the op-eration of the dispatch. And the capacity market we proposedonly takes place when it is required, i.e., in the situation of pos-sible congestions predicted by the DSO.

Besides, we have made some assumptions that FOs fully co-operate in the grid capacity market to avoid the congestion is-sue. This assumption is made based on the discussions in (Mc-Calley et al., 2003; Raiffa, 1982), in which, three types of ne-gotiations have been characterized as either strident antagonist,cooperative antagonist or fully cooperative. The first one meansthe agents are complete distrust each other. The latter meansthe agent are entirely self-interested but ones that recognize andabide by whatever rules exist. The third type means the agentsperform no strategic posturing and think of themselves as a co-hesive entity with intension to arrive the best decision for theentity, although they have different needs, values, etc. By be-ing fully cooperative in our context, FOs will honestly submitthe bids based on their marginal cost functions and the impar-tial market operator will update the market clearing price onlyreflecting the constrained resources (distribution grid capacity).

Lastly, in this shadow price mechanism based method, FO,EV owner needs to pay the higher shadow price if they chargethe EVs in the time slot where congestion happens and DSOseems cost nothing to eliminate the grid congestion. However,in the real practices, it is the DSO’s responsibility to upgradethe network to address the challenges. It is therefore assumedthat the shadow price can be modified when sending to the FOsor the FOs may get compensation from the DSO. In addition tothis, DSO need to support the operation of the market operatorand can investigate the saving on the information and commu-nication infrastructures in the distribution grid.

4

3. Problem formulation and control algorithms develop-ment for EV charging schedule generation and grid con-gestion management

In this section, we firstly introduce the newly defined methodfor the EV charging schedule generation, then summarize thekey elements of the mathematical formula part of our previ-ous work (Hu et al., 2013), i.e., the algorithm for shadow pricebased coordination.

3.1. EV charging schedule generation

Linear programming is used and modified to model thecharging process of EVs (Hu et al., 2013, 2011). The objec-tive is to minimize the charging cost as well as fulfilling thedriving requirement of the EV owner. The scheduling periodis divided into NT time slot and the time slot could be hourlybased or fifteen/ten minutes based depending on the modelingrequirements. The objective function is defined as the prod-uct of virtual price (predicted electricity price and the weightedshadow price, in which the shadow price reflect the congestioncost of the distribution grid) and a decision variable P j,i, wherej = 1, 2, ...,NE

k is the index for the number of EVs under oneFO, NE

k denotes the number of EVs under FO k. i = 1, 2, ...,NT

is the index for the time slot in the scheduling period. The phys-ical meaning of the decision variable P j,i is to make a decisionto distributed/charge the power on the certain time slots, wherethe charging cost can be minimized. Predicted electricity priceis assumed to be known in each time slot. With the defined ob-jective function and the constraints such as: 1) the available en-ergy in the battery should be greater than or equal to the energyrequirement for the next trip/time slot. 2) the available energyin the battery should be less than or equal to the power capacityof the battery. 3) The charging rate should less than or equal toits maximum power rate of the charger, the mathematical modelof the solution is presented as follows:

minimizeNT∑

i=1

(Φ j,i + ξi ∗ Λ(i))P j,it, j = 1, ...,NEk

subject to

S OC0, j +NT∑i=1

P j,it j,i ≥ S OCMin, j +NT−1∑i=0

Edrive,i+1

S OC0, j +NT∑i=1

P j,it j,i ≤ w ∗ Ecap, j +NT+1∑i=2

Edrive,i−1

0 ≤ P j,it j,i ≤ Emax, j, i = 1, ...,NT

(1)

where Φ j,i means predicted day-ahead electricity market pricevector, Λ(i) represents the shadow price, ξi denotes the respond-ing weighting factor the shadow price, t means length of eachtime slot. S OC0, j denotes the initial SOC of individual EV.S OCMin, j denotes the recommended minimum SOC of the EV.Edrive means the predicted individual EV owners driving re-quirement. Emax, j denotes the charge rate in term of energy ofindividual EV. w ∗ Ecap, j means the recommended maximumSOC of the EV, where w is the parameter which express the

charging behavior of the battery of the EV is a linear process,Ecap, j is the capacity of the battery of the EV.

With the above optimization problem, each EV agent cangenerate a unique energy schedule; the sum of the individualEV energy schedule in one FO will be denoted as PE

k,i, and

PEk,i =

NEk∑

j=1

P j,i, k = 1, ...,NB, i = 1, ...,NT ,

where NB represents the number of the FOs, k denotes the indexfor the number of FOs, k = 1, ...,NB.

This is the key computation method that is used in this studyfor step 1 which is described in section 2.3. Then in step 2, 3,there is no issue needed to be clarified. In step 4, Distributionsystem operator pre-access and analyse the distribution networkby running the load flow calculation in the simulink where a10kV distribution network is modeled. The math behind step 5,6, 7 will be explained in the following subsection.

3.2. Price based approach for distribution grid congestionmanagement

To describe the price coordination method, we start with aproposed cost function which represents the cost of the powerpreference difference of a FO in each time slot, e.g.,

µk = ζk(Pk,i).


µk = Ck,i(Pk,i − PEk,i)

2, (2)

where i, k, PEk,i keep the same with above notation, Pk,i denotes

the control variable, Ck,i means the weighting factor which areassociated with the power difference, the larger Ck,i implies asmaller difference. The objective is to minimize the cost func-tions of all the FOs as well as respect to the constraint from theDSO:

minimizeNB∑

k=1

NT∑

i=1

Ck,i(Pk,i − PEk,i)

2

subject to

NB∑

k=1

Pk,i ≤ PCap(i), i = 1, ...,NT , (3)

where PCap(i) is the power capacity specifically for all the FOs,for example, it can be estimated by the DSO after deductingthe conventional loads. This problem is a convex optimizationproblem and relevant research (Boyd and Vandenberghe, 2004;Boyd et al., 2007) show that by introducing Lagrange multipli-ers or shadow price Λ(i) ∈ RNT , problem (3) can be transferredinto following partial Lagrangian problem:

L =

NB∑

k=1

NT∑

i=1

Ck,i(Pk,i−PEk,i)

2 +

NT∑

i=1

Λ(i)(NB∑

k=1

Pk,i−PCap(i))(4)

The centralized optimization problem (3) is transferred intoa decentralized one with associated shadow price Λ(i) in each

5

time slot, with the purpose of emulating the market behavior.In the starting point, the shadow price is assumed to be zero,then the optimal solution for equation (4) are PE

k,i. This explainsthe step 5 that FOs first directly submit their power scheduleto the market operator and market operator will determine theshadow price. Since the market operator’s interest is in alliancewith the DSO, i.e., eliminating the grid congestion, as furtherexplained in study (Hu et al., 2013) and (Boyd et al., 2003), theshadow price can be updated according to Λ(i)ω+1 = Λ(i)ω +

αω.(∑NB

k=1 P∗k,i(Λ∗) − PCap(i)) until the price converges, where

P∗k,i(Λ∗) is the optimal solution of equation (4) with the given

Λ∗, i.e., the newly Λ(i)ω+1, ω is the convergence steps needed,αω ∈ R denotes the step size and can be chosen as αω = αwhichis a positive constant, independent of k; with such choice, theconvergence is guaranteed. This explains the step 6.

In step 7, FOs sends the shadow price to all the EV controller.Then EV controller restarts the step 1, the only difference is thata shadow price is added on the top of the predicted spot energyprices, and the modification compared to the study (Hu et al.,2013) lies on the response weighting factor γ to the shadowprice. The EV owner can show their will by setting up this re-sponse weighting factors. For example, if γ is zero, it representsthat the EV owner is fully insensitive to the shadow price andwill keep the original power schedule; otherwise, a new powerschedule will be generated and submitted to the FOs. By re-peating the steps, the proposed solution can ensure the safety ofthe grid in the planning period.

4. Multi-agent model for control system demonstration

4.1. Multiagent system architectureIn order to demonstrate the operation of the control systems,

a multi-agent system is utilized and developed in this study. Fig.3 depicts the MAS system architecture, in which, all the agentsare built in JACK which is an agent-oriented development envi-ronment built on top of and fully integrated with the Java pro-gramming language (Howden et al., 2001). JACK offers theenvironment and facilities message sending/receiving. Matlabbased functions enables a declarative implementation of the de-cision module. Simulink is used to model the distribution gridand functioned for power flow calculation. The java applica-tion programming interface matlabcontrol 2 is used for JACKto interact with Matlab.

4.2. Introduction on the features of JACKThe agents used in the JACK are modeled according to the

theoretical Belief Desire Intention (BDI) model of artificialintelligence (Wooldridge, 2008). Within the environment, aJACK agent is a software component that can exhibit reason-ing behaviour under both pro-active (goal directed) and reactive(event driven) stimuli. As key components of JACK, the JACKagent language introduces five main class-level constructs:

• Agent: which models the main reasoning entities in JACK.

2https://code.google.com/p/matlabcontrol/

FOi agent

DSO agentEV agent-n

Market operator

agentMulti Agents

coordination layerJACK

MULTIAGENTS SYSTEM

FOi

optimization slover for market

participationMatlab based

EViSchedule

generationMatlab based

Conventional load

EVn

Market operation

Matlab based

Grid SimulationSmulink

Optimization tools Matlab

Figure 3: A MultiAgent system architecture.

• Event: which models occurrences and messages that theseagents must be able to respond to.

• Plan: which models procedural descriptions of what anagent does to handle a given event. An agent’s plans areanalogous to functions.

• Capability: which aggregates functional components(events, plans, beliefsets and other capabilities) for agentsto make use of.

• Beliefset: which models an agent’s knowledge about theworld.

In this study, we mainly use the three class levels: the agentclass, the event class, and the plan class.

4.3. Use case and the MAS implementation

4.3.1. Agent class and its instantiationConsidering the features of JACK and the requirements of

our desired systems, we use agent class in the design viewswhich make the instantiation of an agent class flexible. Be-sides, each agent has several plans which are used to handle theevents.

EV agent: An EV agent class is responsible for generatingthe charging schedule of individual EVs. They communicatewith the subscribed FOs.

FO agent: A FO agent class is responsible for aggregatingthe charging schedule of their contracted EV agents and mod-ifying the power schedule when negotiating with the marketoperator agent. In short, the FO agent communicates with EVagents, the DSO agent and the market operator agent.

DSO agent: A DSO agent is responsible for the grid safetyby performing load flow calculation after obtaining the power

6

DSOMarket

Operator


NotifyCongestion

ResponseCongestion

FO_PowerSchedule

ShadowPricetoFO

MarketOperation

FO_ScheduleAD

SendPowerSchedule

FO

FO_PowerScheduleAggreation

handles

handles

sends

handles

handles

handles

handles

handles

handles

handles

sendssends

sends

sends

sends

uses

uses

uses

uses

EV

SelfPostInformation

ChargingSchedule AskingPowerCalculation

EV_SendChargingSchedule

EVSelfInformation

EV_ChargingSchedulePreparing

FO_CalculationCenter

PriceCenterToEV

FinalShadowPrice

Agent

Plan

Event

handlesposts

handles

uses

sends

handles

uses

uses

sends

handles

handlesuses

sends

handles

handles

sends

sends

sends

sends

sends

Figure 4: The diagram of JACK implementation.

schedules of FOs. The DSO agent communicates with the FOsagent and the market operator agent.

Market operator agent: A market operator agent is respon-sible for the making of the shadow price. The market operatoragent communicates with the DSO agent and the FOs agent.

4.3.2. A multiagents system built on JACKFig. 4 shows the whole design diagram for the desired multi-

agent systems in the JACK, which is built according to the pro-posed solutions in this study, i.e., the eight steps presented in theabove. We will explain this diagram according to the sequenceof the steps and divide it into three parts. Besides, the contentinside each box of this diagram (mainly the content inside theplans of each agent) will be explained.

• The interaction between the EV agent and the FO agent.

In the implementation, we define that the FO provides thecalculation center to the EV agents to facilitate the com-putation, although it is assumed that EV agent makes thecharging schedule by himself/herself. With this set up, theprogramming time can be saved significantly.

Event SelfPostInformation: Posted by the EV agent, thepurpose is to trigger the plan EVSelfInformation.

Plan EVSelfInformation: The EV agent read the infor-mation including initial SOC, the driving requirement ofthe EVs in the scheduling period, the bus information 3

3Buses of the modelled distribution system in the Simulink, in which severalload buses are defined for connecting the EVs.

and the response weighting factor to the shadow price. Af-ter obtaining the personal information, the EV agent sendsan event named AskingPowerCalculation to the FO agent,and the event will be handled by the plan FOCalculation-Center.

Plan FOCalculationCenter: An Matlab based programwill be called and be used to generate the charging powerschedule. The power schedule will be sent back again tothe EV agent by using the event ChargingSchedule.

Plan EVChargingSchedulePreparing: Send the powerschedule and the corresponding bus information to theFOs agent by event named EVSendChargingSchedule, thisevent will be handled by the plan FOPowerScheduleAg-gregation.

Plan FOPowerScheduleAggregation: All the contractedEV’s power schedule will be summed according to whichbus they are connected.

• The interaction between the FO agent and the DSO agent

Event SendPowerSchedule: Each FO agent sends the ag-gregated power schedule to the DSO agent by this event.The event will be handled by the DSO agent with the planVerifyGridCongestion.

Plan VerifyGridCongestion: The DSO agent will call thegrid model built in the simulink with the newly powerschedule of the FOs and the conventional loads and do thepower flow calculation. The DSO agent can fetch the valuefrom the simulink and compare it with the capacity of thedistribution grid, such as the transformers. Then the DSOagent will send the result to all the FO agents through theevent named NotifyCongestion. The event will be handledby the plan ReponseCongestion.

Plan ReponseCongestion: All the FO agents get the resultand check whether congestion exists. If it is congested, allthe FO agents will resort to the market operator agent tonegotiate the power capacity. Otherwise, the FO agentsare allowed to bid the energy schedule into the energy spotmarket.

• The interaction between the FO agent and the Market op-erator agent

Event FOPowerSchedule: FO agent sends the powerschedule to the market operator agent by this event andthe event will be handle by the market operator agent withthe plan MarketOperation.

Plan MarketOperation: In the plan, the market operatoragent calls the matlab based price determination programand check whether the price is converged every iteration.If the price is not converged, the market operator agentwill send the updated shadow price to the FO agent by theevent ShadowPricetoFO. Accordingly, this event will behandled by the agent with the plan FOScheduleAD.

Plan FOScheduleAD: In the plan, the FO agent willreschedule the power based on the predefined cost func-tions and the updated shadow price and resend the power

7

B21B22

B3

B31B32

B4

B41B42

B5

L1L2

Line4

B1

B2

Transformer60 kV/11 kV

External grid

Line5 Line1

Line2Line6Line7

Line8Line9

Line3

L3L4

L5L6

L7

T1

Potential overloading area.

Figure 5: A 10kV distribution network.

schedule to the market operator agent by the some eventFOPowerSchedule. If the price is converged, the marketoperator agent will send the final shadow price to the FOagent by the event FinalShadowPrice. The event will behandled by the FO agent with the plan PriceCenterToEV.

Plan PriceCenterToEV: In the plan, the FO agent doesnot sends the shadow price to the EV agent directly, be-cause the calculation center is place on the FO agent level.Instead, it sends a normal message which stimulate theevent SelfPostInformation again.

Note that when the price is converged, a complete se-quence of the operation needed for grid congestion hasbeen went through. However, the newly acceptable powerschedule of the FOs might deviate the original plan. There-fore, we leave the chance to the FOs and the EV owners tomake a new schedule based on the information of the firstround. This explain the fact of sending the final shadowprice to the FO agent by using the event FinalShadow-Price. In order to run the complete sequence again, the EVagents need a signal to stimulate the corresponding plansand this is achieved by the event SelfPostInformation sentout by the market operator agent using the plan PriceCen-terToEV.

5. Simulation and demonstration results

5.1. Case study specification

A 10kV radial network is considered in this case study, theone-line diagram/topology of the network is shown in Fig. 5.The network is modified from Østergaard and Nielsen (2008);Han (2012), which can represent the typical features of a Dan-ish distribution systems. The network consists of two voltage

level, 11 buses, 9 distribution lines, 7 load buses and the net-work is modeled in the Simulink. 1400 households is connectedin this distribution systems, and 20% of the households is as-sumed to have EVs. Considering the similarities of the driv-ing patterns of the EV users and simulation requirements ofthe multiagent systems, we divide the 280 EVs into 14 groupswhich is represented by 14 EV agents. Three FOs is assumedto provide services to these 14 EV agents. FO1 is responsiblefor EV agents EV1 to EV5. EV6 to EV9 is assigned to FO2.The rest of the EV agents subscribe to FO3. If all the EVs isconnected to the grid at the same time, this will bring 644kWadditional load to the network (Maximum individual EV charg-ing rate is limited to 2.3 kW, this fits with the Danish case (10 A,230 V connection)). In our case study, we set up the availablepower capacity for all the EVs is 600kW (available capacity ofthe primary transformer for EVs ). The weighting factor rateC1,i,C2,i,C3,i is set to 0.5, 0.1, and 0.2. The value of αω is cho-sen as 0.1 in this case.

Figure 6: EV energy driving requirement per EV agent per FO.

For the EV charging schedule, the information of hourlyelectricity spot price of the Nordic power market 4 is assumedto be perfectly known by the EVs, and the price data is identicalwith previous study (Hu et al., 2013). The artificial driving dataof the 14 EV agents have been generated based on the 2003AKTA Survey [31], in which 360 cars in Copenhagen weretracked using GPS from 14 to 100 days. Each data file includesstarting and finishing time, and the corresponding duration anddistance. The original data is transferred into 15 minutes in-terval driving energy requirements based on the assumption of11 kWh/100 km. The 15 minutes interval is changeable ratherthan absolute. The energy driving requirement of EV1 to EV14is illustrated in Fig. 6. It is seen from Fig. 6 that most EVs havea regular pattern, i.e., they leave home in the morning time andcome back in the evening time, while some EVs have higherenergy driving requirement, such as EV agent EV13 which is

4http://www.nordpoolspot.com/

8

shown by the green curve of the bottom figure. For other pa-rameters:

• Battery capacity of all the EV agents are set to 20 kWh.

• Initial SOC of all the EV agents are set to 0.2 of the batterycapacity.

• Minimum SOC of all the EV agents are set to 0.2 of thebattery capacity.

• Maximum SOC of all the EV agents are set to 0.85 of thebattery capacity, the minimum and maximum SOC set upis to ensure that the EV charging process is linear.

• EV agents’s responding weighting factor to the shadowprice is assumed to be (0.01, 0.01, 0.01, 0, 0, 0.01, 0.01, 0,0, 0.01, 0.01, 0.01, 0, 0), correspondingly.

5.2. Simulation results in MATLAB

In this simulation part, we compared the result of two caseswhere the DSO both use the price-oriented market protocolsto interact with the FOs, however, the coordination methodsbetween the FOs and the EVs are different. In the first case, weassume that three FOs centrally schedule and control the EVcharging which is the scenario in work (Hu et al., 2013), whilein the second case, it is assumed that three FOs only aggregatethe charging schedules which are made by the EV controllers,this is the scenario in this study. As illustrated in Fig. 7, thecongestion problems are solved after 5 steps in the first casewhile only 2 steps in the second case. The difference is becausethat the EVs in the first case are always responding the shadowprice and trying to avoid the charging on the higher price period,as a result, the EVs will be scheduled to charge at other lowerprice period where congestion might happens as well. While inthe second case, only some EVs are assumed to responds to theshadow price which means that only part of the charging plan isrescheduled to other lower price period and thereby reduce thepossibility of causing a new congestion period. Note that in thebeginning, the shadow price is zero, so the blue curves in theleft part of Fig. 7 represent the spot electricity price. For therest of the price curves, the spikes represents the shadow prices.

5.3. Demonstration result of the MAS

When setting up the demonstration of the MultiAgent sys-tem, the Simulink part is not included in this case study becausethe capacity limit is only considered for the transformer. Con-sidering the two assumptions: 1) there is no power losses in thedistribution network or we do not consider it. 2) the overheadline and underground cable are capable of handling the increas-ing loads, the power information below the transformer can besimply obtained by summing up the power schedule of the FOsinstead of fetching it from the Simulink. For the rest of thesystem, it goes the same as we presented in Fig. 4. In JACK,there are a number of tools available to assist a detailed trace ofthe system execution which range from graphical tracing toolsto logging tools. In this study, we run the program with theinteraction diagram. As we have one DSO agent, one market

operator agent, three FO agents, and fourteen EVs agents, theinteraction diagram which shows the communication messageamong them is quite large. It is not wise to show the wholeinteraction diagram in this paper, instead, we only show partof the interaction diagram where the message sequence hap-pens between the DSO agent, the market operator agent namedCMO, the FO agent FO1, and one EV agent EV1, this is shownin Fig. 8. The sequence diagram starts from agent EV1 (holdsfor other 13 EV agents) with a request of schedule calculation.Then the schedule information is aggregated by the FO agentand is sent to the DSO agent. The rectangular box marked withiteration represents the interactions between the market oper-ator agent and the FO1 agent. It well emulates the negotia-tion behavior inside a capacity market. When the shadow priceis converged, the shadow price is sent to the EV agent. Withthe new schedule, it is confirmed by the DSO agent that therewill be no congestion for the grid in the planning phase, whichmeans the programming stops.

DSO CMO FO1 EV1

Schedule calculation()

Schedule

Schedule, bus

Checking Congestion()

Checking result

power schedule

Shadow price

IterationIteration Five iterations

Shadow price

Schedule calculation()

Schedule

Schedule, bus

Checking Congestion()

Checking result

No congestionNo congestion Programming stop

Figure 8: Sequence diagram between the chosen agents.

9

(a) The sum of the spot price and shadow price in each iteration step. (b) The comparisons of FO’s power schedule in each step with the power ca-pacity.

(c) The sum of the spot price and shadow price in each iteration step. (d) The comparisons of FO’s power schedule in each step with the power ca-pacity.

Figure 7: Top: Case study for centralized control between FOs and EVs. Bottom: Case study for decentralized control between FOs and EVs.

6. Discussion and conclusion

A multi-agent system is developed to demonstrate the dis-tributed implementation of the grid congestion managementscheme of distribution network with a large scale of EVs. Itis learned from the experience that the distribution grid con-gestion can be eliminated according to economical principles,and a MAS based distributed implementation is of higher ad-vantage. In this study, we develop and utilize an integrated en-vironment consisting of JACK agent software and Matlab toanalyze the cyberphysical aspects of the environment. Thisis because JACK is good for demonstrating the coordinationschemes among the actors, and Matlab is good for technicalcomputation of optimization problem. For a general case, var-ious simulation platform can be utilized in a distribution gridcongestion demonstration. For example, besides JACK, JADEis also widely used for multiagent simulation. We choose JACKbecause of its capability and support for the explicit modelling

of the typical MAS entities such as agent, plan, event and ca-pabilities. Moreover, in JCK it is easier to design and analysisinteractions and dependencies among such entities. In term ofsolving an optimization problem, GAMS also has good perfor-mance, however, Matlab is more widely used in the academi-cally field. Last but not least, grid modeling tool is also an im-portant part, the currently existing grid modeling tools includeSimulink, MatPower, PowerFactory, ARISTO, NEPLAN etc.The scope of these listings is not to give a comparison aboutthe various platforms, instead, we want to emphasize that therelevant tools can be integrated with the MAS settings.

Besides EVs, some other new loads such as heat pumps andthe increasing electrification of the loads in the home will alsobring challenges to the distribution grid. We believe that thismultiagent framework can be used to address the similar chal-lenges. Since the FOs (You, 2010) (Alternative names for an FOare used such as virtual power plant, aggregator) is also widely

10

proposed for aggregating other distributed energy resources. Asexpected, the FOs will represent the DERs and interact withthe market operator and the DSO similarly with the one in thisstudy.

Acknowledgment

The authors would like to thank Nicholas Honeth from RoyalInstitute of Technology for his help on JACK implementation.Also, the authors are grateful to the financial support of theDanish iPower project.

References

Akkermans, H., Schreinemakers, J., Kok, K., 2004. Microeconomic distributedcontrol: theory and application of multi-agent electronic markets, in: Pro-ceedings of CRIS, pp. 163–176.

Andersen, P., Hu, J., Heussen, K., 2012. Coordination strategies for distributiongrid congestion management in a multi-actor, multi-objective setting. IEEEInnovative Smart Grid Technologies Conference Europe , 1–8.

Bessa, R.J., Matos, M.A., 2012. Economic and technical management of anaggregation agent for electric vehicles: a literature survey. European Trans-actions on Electrical Power , 334–350.

Boyd, S., Vandenberghe, L., 2004. Convex optimization. Cambridge UnivPress.

Boyd, S., Xiao, L., Mutapcic, A., 2003. Subgradient methods. Stanford Uni-versity, Autumn Quarter 2004.

Boyd, S., Xiao, L., Mutapcic, A., Mattingley, J., 2007. Notes on decompositionmethods. Stanford University .

Cheng, J.Q., Wellman, M.P., 1998. The walras algorithm: A convergent dis-tributed implementation of general equilibrium outcomes. ComputationalEconomics 12, 1–24.

Christie, R.D., Wollenberg, B.F., Wangensteen, I., 2000. Transmission manage-ment in the deregulated environment. Proceedings of the IEEE 88, 170–195.

Clement-Nyns, K., Haesen, E., Driesen, J., 2010. The impact of charging plug-in hybrid electric vehicles on a residential distribution grid. IEEE Transac-tions on Power Systems 25, 371–380.

Green II, R.C., Wang, L., Alam, M., 2011. The impact of plug-in hybrid electricvehicles on distribution networks: A review and outlook. Renewable andSustainable Energy Reviews , 544–553.

Greunsven, J., Veldman, E., Nguyen, P., Slootweg, J., Kamphuis, I., 2012. Ca-pacity management within a multi-agent market-based active distributionnetwork, in: 2012 3rd IEEE PES Innovative Smart Grid Technologies, IEEE.pp. 1–8.

Han, X., 2012. Quantitaive analysis of distribtued energy resources in futuredistribution networks. Master’s thesis. Royal Institute of Technology.

Heydt, G.T., 1983. The Impact of Electric Vehicle Deployment on Load Man-agement Strategies. IEEE Power Engineering Review PER-3, 41–42.

Howden, N., Ronnquist, R., Hodgson, A., Lucas, A., 2001. Jack intelligentagents-summary of an agent infrastructure, in: 5th International conferenceon autonomous agents.

Hu, J., You, S., Lind, M., Østergaard, J., 2013. Coordinated charging of electricvehicle for congestion prevention in the distribution grid. accepted for thepublication on IEEE Transactions on smart grids .

Hu, J., You, S., Østergaard, J., Lind, M., Wu, Q., 2011. Optimal chargingschedule of an electric vehicle fleet, in: Proceedings of 2011 46th Interna-tional conference on Universities’ Power Engineering Conference (UPEC),,VDE. pp. 1–6.

Issicaba, D., Rosa, M., Franchin, W., Lopes, J.P., . Agent-based system appliedto smart distribution grid operation .

Jennings, N.R., Bussmann, S., 2003. Agent-based control systems. IEEE con-trol systems 23, 61–74.

Kabisch, S., Schmitt, A., Winter, M., Heuer, J., 2010. Interconnections andcommunications of electric vehicles and smart grids, in: 2010 First IEEE In-ternational Conference on Smart Grid Communications (SmartGridComm),IEEE. pp. 161–166.

Karfopoulos, E.L., Hatziargyriou, N.D., 2013. A Multi-Agent System for Con-trolled Charging of a Large Population of Electric Vehicles. IEEE Transac-tions on Power Systems 28, 1196–1204.

Lopes, J.A.P., Soares, F.J., Almeida, P.M., da Silva, M.M., 2009. Smart Charg-ing Strategies for Electric Vehicles: Enhancing Grid Performance and Max-imizing the Use of Variable Renewable Energy Resources, in: EVS24-The24th International Battery, Hybrid and Fuel Cell Electric Vehicle Sympo-sium & Exhibition, Stavanger, Norway.

Lopes, J.A.P., Soares, F.J., Almeida, P.M.R., 2011. Integration of electric vehi-cles in the electric power system. Proceedings of the IEEE 99, 168–183.

Lorenz, G., Gonzalez-Gutierrez, C., Schlosser, P., 2009. Dsos as market facili-tators .

Lyon, T.P., Michelin, M., Jongejan, A., Leahy, T., 2012. Is smart chargingpolicy for electric vehicles worthwhile? Energy Policy 41, 259–268.

Ma, Z., Callaway, D.S., Hiskens, I.A., 2013. Decentralized Charging Controlof Large Populations of Plug-in Electric Vehicles. IEEE Transactions onControl Systems Technology 21, 67–78.

McCalley, J., Zhong, Z., Vishwanathan, V., Honavar, V., 2003. Multiagentnegotiation models for power system applications. Autonomous systemsand intelligent agents in power system control and operation, C. Rehtanz,editor, Springer-Verlag, Berlin , 49–74.

Miranda, J., Borges, J., Mendes, M.J., Valerio, D., 2011. Development of amulti-agent management system for an intelligent charging network of elec-tric vehicles, in: 18th IFAC World Congress, pp. 12267–12272.

Nagata, T., Sasaki, H., 2002. A multi-agent approach to power system restora-tion. IEEE Transactions on Power Systems 17, 457–462.

Nordentoft, N.C., 2013. iPower WP3.8 Development of a DSO-market on flex-ibility services. Technical Report. iPower Consortium.

Nordman, M.M., Lehtonen, M., 2005. An agent concept for managing electricaldistribution networks. IEEE Transactions on Power Delivery 20, 696–703.

O’Connell, N., Wu, Q., Ostergaard, J., Nielsen, A., Cha, S.T., Ding, Y., 2011.Electric vehicle (ev) charging management with dynamic distribution sys-tem tariff, in: 2011 2nd IEEE PES on Innovative Smart Grid Technologies,pp. 1–7.

Østergaard, J., Nielsen, J.E., 2008. The bornholm power system an overview.Pipattanasomporn, M., Feroze, H., Rahman, S., 2009. Multi-agent systems in

a distributed smart grid: Design and implementation, in: 2009. PSCE’09.IEEE/PES Power Systems Conference and Exposition, IEEE. pp. 1–8.

Raiffa, H., 1982. The art and science of negotiation. Harvard University Press.Rehtanz, C., 2003. Autonomous systems and intelligent agents in power system

control and operation. Springer.Ren, F., Zhang, M., Sutanto, D., 2013. A multi-agent solution to distribution

system management by considering distributed generators. IEEE Transac-tions on Power Systems 28, 1442–1451.

Richardson, P., Flynn, D., Keane, A., 2012. Local versus centralized chargingstrategies for electric vehicles in low voltage distribution systems. SmartGrid, IEEE Transactions on 3, 1020–1028.

San Roman, T.G., Momber, I., Abbad, M.R., Sanchez Miralles, A., 2011. Regu-latory framework and business models for charging plug-in electric vehicles:Infrastructure, agents, and commercial relationships. Energy Policy 39.

Schlosser, P., 2010. The Role of Distribution System Operators (DSOs) asInformation Hubs. Technical Report. Eurelectric.

Schmutzler, J., Wietfeld, C., 2010. Analysis of message sequences and en-coding efficiency for electric vehicle to grid interconnections, in: VehicularNetworking Conference (VNC), 2010 IEEE, IEEE. pp. 118–125.

Shao, S., Zhang, T., Pipattanasomporn, M., Rahman, S., 2010. Impact of TOUrates on distribution load shapes in a smart grid with PHEV penetration.2010 IEEE PES Transmission and Distribution Conference and Exposition:Smart Solutions for a Changing World .

Søndergren, C., 2011. Edison project: Electric Vehicles in Future Market Mod-els. Technical Report. Danish Energy Association.

Sundstrom, O., Binding, C., 2012. Flexible Charging Optimization for ElectricVehicles Considering Distribution Grid Constraints. IEEE Transactions onSmart Grid 3, 26–37.

Wellman, M.P., 1993. A market-oriented programming environment and its ap-plication to distributed multicommodity flow problems. Journal of ArtificialIntelligence Research 1, 1–23.

Wooldridge, M., 2008. An introduction to multiagent systems. Wiley. com.You, S., 2010. Developing Virtual Power Plant for Optimized Distributed En-

ergy Resources Operation and Integration. Ph.D. thesis. PhD thesis, Techni-cal University of Denmark, 2011.

11


A.10 HV/MV power transformer capacity ne-gotiation by fleet operators using multi-agents

The main content of this paper was presented at International Conference onLife System Modeling and Simulation, 2014, Shanghai.

adfa, p. 1, 2011.

© Springer-Verlag Berlin Heidelberg 2011

HV/MV Power Transformer Capacity

Negotiation by Fleet Operators using Multi-agents

Junjie Hu, Hugo Morais, Morten Lind, Jacob Østergaard

Department of Electrical Engineering / Technical University of Denmark (DTU)

Elektrovej, Bld 325, 2800 Lyngby, Denmark

{morais, junhu, mli, joe}@elektro.dtu.dk

Abstract. The integration of electric vehicles in future power systems will

change their normal operations. The system operators will need to manage this

new energy resource, considering the charge necessities as well as the discharge

energy opportunities. Congestion situations can occur in some equipment’s

(lines and power transformers), mainly in peak periods. This paper proposes a

hierarchical management structure considering the distribution system operator

and a new entity called electric vehicles fleet operator. To simulate this collabo-

rative (all players contribute to the operation system stability) but also competi-

tive environment (each player will try to increase its profits or reduce its costs),

a multi-agent platform was developed to demonstrate the interactions between

the entities.

Keywords: Congestion Management, Electric Vehicles, Fleet Operators, Multi-

agent Systems, Smart Grids

1 Introduction

Several policies were issued in recent years to improve the energy consumption ef-

ficiency, to change the energy mix and to reduce the greenhouse gas emissions. The

incentives provided to distribution generation (DG) have been a success and in sever-

al countries, DG represents more than 20% in the electricity generation mix. The wind

generators and solar photovoltaic panels are the technologies with higher growing.

However, technologies such a micro-hydro generation or combined heat and power

(CHP) represent important energy resources in several countries.

Despite the high penetration of DG technologies the governments in European

countries defined ambitious goals for a near future. According the European energy

roadmap 2050 [1] “The EU is committed to reducing greenhouse gas emissions to 80-

95% below 1990 levels by 2050 in the context of necessary reductions by developed

countries as a group”. In some countries, the goals are even more challenging. For

example in Denmark, in 2020 the Danish government intends a reduction of 12% of

energy consumption (in comparison with consumption in 2006) and also supplies

50% of the electricity consumption using wind generation [2]. In 2035 the Danish

government intends to have all the power systems based on renewable resources and

in 2050 intends to have all the economy, including the power systems, the heating

systems and also the transportation, based in renewable resources [3].

To achieve these objectives it is necessary more investment in DG but also in in-

frastructures (transmission and distribution networks) and in new management meth-

odologies in order to assure the reliability in power systems operation. The use of

storage units like batteries, pumped-hydro plants, compressed air energy storage or

other technology is crucial for the future power systems based in renewable and in-

termittent resources.

In another point of view, to guarantee the greenhouse gas (GHG) emissions reduc-

tion it is necessary the decarbonisation of all activities, besides the power industries.

The power industries were responsible for around 30% of GHG in EU-27 in 2011.

The second sector with more GHG emissions in 2011 were the transports with a share

of 20.3% [4]. To assure the GHG emissions reduction in transports and also in heat-

ing/cooling it is necessary the use of electricity in these sectors. The use of electric

vehicles will be essential to assure the GHG emission reduction goals. In the end of

2012 had been sold more than 180 000 electric vehicles (EVs) in worldwide and are

expected a total of 20 million on the road in 2020 [5].

The integration of these vehicles in the electric system will increase significantly

the global power demand [1]. In this sense it is necessary develop new energy re-

sources management strategies mainly in distribution level avoiding congestions in

the network. EVs can be managed as a distributed energy resources providing a flexi-

ble storage system in low voltage distribution network [6]. EVs can be used as reserve

providing different kind of ancillary services helping the system operators maintain-

ing the system stability [7]. EVs are very flexible being potentially capable of enhanc-

ing the efficiency of other DERs. However, uncontrolled charging of EV can create

new load peaks during the day, increasing power losses, the voltage deviations and

the network congestion [8].

EV fleet operator (FO) is a new entity aiming to capture the business opportunities

by providing the multiple services of EVs. EV FO could be independent or integrated

in an existing business function of the energy supplier. EV FO intends to guarantee

driving needs of the EV owners, coordinate and support the valuable services and

operation constraints of EV and power system operator, maximize the renewable

energy use, implement centralized control/marketing method to maximize business

values, optimize the EVs charging and discharging processes [9].

In the present paper a multi-agent platform implemented to simulate the interac-

tions between the EV FOs and the distribution system operator (DSO) and between

the EV FOs and the EVs owners is presented. The main goal is the negotiation be-

tween the agents in order to avoid the congestion of the distribution network lines and

mainly the HV/MV power transformer. Each EV FO has the capability to manage the

EVs charge and discharge considering the energy prices and the EVs requirements

(schedule trips, batteries technical limits, etc). The problem considers the network

technical constraints namely the bus voltage magnitude and the lines thermal limits.

This paper is organized as follows: after this introductory section, section 2 pre-

sents a description about the fleet operators functioning, section 3 presents the con-

gestion negotiation method, section 4 describes the implemented multi-agents plat-

form and the negotiation mechanism. The main conclusions of the paper are provided

in section 5.

2 Fleet Operators

Fleet Operators (FO) can have different definitions according the context. In this

specific case the FO can be described as the entities responsible for the electric vehi-

cles (EVs) and plug-in hybrid electric vehicles (PHEV) charge and discharge coordi-

nation [10]. The use of electric vehicles with gridable capability, normally called

vehicle to grid (V2G) has been discussed in several papers [6], [11], [12] considering

different management methodologies. The main conclusion of the EVs integration in

distribution network studies and analysis is that the electric vehicles charge and dis-

charge should be coordinated in order to avoid critical situations mainly during the

peak periods. In fact the peak consumption can increase significantly if the all EVs

charge at the same time. On the other hand, EVs can provide some interesting support

to the distribution network operation functioning as small batteries capable to dis-

charge energy to the network when the system operator has need.

Aggregators like virtual power plants or fleet operators will be crucial in a near fu-

ture in order to coordinate the EVs charge and discharge process. The main goal is try

to schedule the EVs charge during the off-peak periods and, when required, discharge

energy in order to support the system operator in the distribution network manage-

ment. Virtual Power Plants are players with the capability to manage several distrib-

uted energy resources technologies like distributed generation units, demand response

programs, storage systems and electric vehicles [13], [14]. FOs is more focus in the

EVs coordination considering the constraints imposed by the DSO, taking into ac-

count the network limitations and also the market opportunities in order to take ad-

vantages in the energy negotiation. In some situations FOs can also manage the loads

consumption. In Fig. 1, the proposed architecture considering two FOs is presented.

FO1

Fleet Operation

Electricity

Spot Market

Grid

Measurements

DSO


Conventional

load

FO2

Fleet Operation

Market

operator

Distribution

grid capacity

market

Information

flow

Control

relation

Physical

connection

Fig. 1. A schematic of a low voltage active distribution system [9]

Fig. 1 presents a schematic of a low voltage distribution system, considering the

connection of electric vehicles and the consumers. In this distribution system, it is

assumed that the consumers own controllable devices, i.e., EVs, besides some con-

ventional loads, such as light or TV. These EVs are divided into two groups as illus-

trated in Fig. 1. One group is controlled by fleet operator-1 (FO1), another group is

controlled by fleet operator-2 (FO2). In this hierarchical distribution system, both FOs

can schedule and control their customer’s electricity consumption directly. While on

the FO level, the coordination between FOs and DSO is made through the distribution

grid capacity market. The capacity market is managed by the DSO allowing the use of

distribution network in a competitive environment mainly in the peak periods. FOs

can also negotiate energy in electricity markets and in bilateral negotiations with other

agents [9].

The coordination between DSO and FOs is crucial to manage the distribution net-

work use avoiding congestion situations [15], [16]. The congestion situations can

occur in the lines but also in the power transformer. In the proposed methodology

both type of congestions are considered. An AC Power flow is included in the FOs

scheduling in order to avoid the congestion in the distribution network. However the

use of power transformer should be managed by the DSO using a negotiation mecha-

nism in order to allocate the power transformer to different FOs.

Each FO will do its EVs charge/discharge scheduling taking into account the net-

work constraints. The scheduling is implemented as a mixed-integer non-linear prob-

lem (MINLP) trying to minimize the costs (1). The energy cost depends of the exter-

nal suppliers as well as by the HV/MV power transformer capacity use. Each FO can

use the energy stored in the EVs batteries avoiding congestion problems. The power

losses in the distribution network are also considered in the problem. Expression (1)

represents the implemented objective function.

2

2

, ,2 1 1

1

, , , ,2 1 1

min

V G l

V G l

N N

Ch Dch Load CongEV EV t l t tTV G l

N Nt

Dch Dch NSD NSDEV t EV t l t l tV G l

P P P c

Z

P c P c

(1)

In expression (1), the variables ,Ch EV tP and ,Dch EV t

P represent the charge and

discharge power of electric vehicles and the parameter ,Dch EV tc the energy dis-

charge cost. The loads are characterized by the consumption forecast ,Load l tP

and by the non-supplied demand ,NSD l tP . This parameter will be important to

avoid the congestion situations in extreme situations, considering the cost

,NSD l tc . ,NSD l t

c can represent the cost with a demand response event or a pe-

nalization to load curtailment without any coordination or contract. The parame-

ter Cong tc represents the HV/MV power transformer congestion cost. This pa-

rameter is different in each negotiation iteration. In the first iteration Cong tc is

zero, representing the situation without congestion.

The problem constraints considers the first Kirchhoff law relating to the active (2)

and reactive (3) power balance considering that the energy is supplied by external

suppliers SP and the distribution network technical limits regarding to the lines ther-

mal limits (4) and the bus voltage magnitude limits (5) The implemented AC power

flow model is based in [17]. In expressions (2) to (5), the parameters i and j represent

the bus i and j. Additionally, the problem considers the maximum power limit im-

posed by DSO regarding the HV/MV power transformer capacity use. The EVs con-

straints relating to the batteries energy capacity limits (6), the batteries internal energy

balance (7) and the charge (8) and discharge (9) rates are also considered [18].

Active Power Balance

( , ) ( , ) ( , ) ( , ) ( , )

1 1 1

2

( ) ( ) ( ) ( ) ( )

( ) ( ) ( )

cos sin

1,..., ; 1,..., ;

i i iSP V L

i

N N Ni i i i i

SP SP t Dch EV t Ch EV t Load L t NSD L t

SP V L

ii i t i t j t ij ij t ij ij t

j L

B ij t i t j t

P P P P P

G V V V G B

t T i N

(2)

Reactive Power Balance

( , ) ( , ) ( , )

1 1

2

( ) ( ) ( ) ( ) ( )

( ) ( ) ( )

sin cos

1,..., ; 1,..., ;

i iSP L

i

N Ni i i

SP SP t Load L t NSD L t

SP L

i t j t ij ij t ij ij t ii i t

j L

B ij t i t j t

Q Q Q

V V G B B V

t T i N

(3)

Lines thermal limits

*

( ) ( ) ( ) _ ( )

*

( ) ( ) ( ) _ ( )

1,..., ; , 1,..., ; ; 1,...,

max

i t ij i t j t sh i i t Lk

max

j t ij j t i t sh j j t Lk

B K

U y U U y U S

U y U U y U S

t T i j N i j k N

(4)

Bus voltage magnitude limits

( ) ; 1,.., i i

Min i t MaxV V V t T (5)

Energy stored limits in EVs batteries in each period t

( , )_ , _ ,Stored EV tSt Min EV t St Max EV tE E E (6)

Energy balance in EVs batteries in each period t

( , ) ( , 1) ( , ) ( ) ( , ) ( , )

( )

1Stored EV t Stored EV t Trip EV t c EV Ch EV t Dch EV t

d EV

E E E P P

(7)

Charge and discharge rates in each period t

( , ) _ ,Ch EV t Ch Max EV tP P (8)

( , ) _ ,Dch EV t Dch Max EV tP P (9)

3 HV/MV Power Transformer use negotiation

The distribution networks have normally the capacity to accommodate all the con-

nected loads considering the consumption evolution for next couple of years. Howev-

er, a large penetration of EVs will bring some challenges to the distribution system

operators or utilities. The challenges usually include peak power issue, grid conges-

tion problem, power losses, voltage drop et al. Much research has been performed to

study the intelligent EV load control and their effect on the grid, which can be dated

back to the early 1980s [19]. In [19] the author argued that load management should

be deployed to alleviate peak loading, which is measured in term of load factor im-

provement. Even low penetration levels of EVs can create new peak loads exceeding

the natural peak if sufficient attention is not paid to distribute the charging load

throughout the off-peak period [20]. A penetration level of 20% is found to be the

upper limit which could be managed by distributing the charging load. Basically,

those studies mainly investigated the impacts by adding the new EV loading profile to

the already existing load profile and seeing the overall effect and then proposed the

load shifting strategy.

Figure 2 presents a much advanced control strategy to prevent the grid congestions.

In the first step, each fleet operator performs the EVs charge and discharge scheduling

considering the inexistence of any congestion in HV/MV power transformer. Howev-

er, a congestion situation inside the distribution network can be avoided due to the

inclusion an AC power flow in the optimization problem constraints. Each FO sends

the initial schedule to the distribution system operator in order to validate the initial

proposals. If the limit of power transformer is not achieved, the DSO approves the

proposals and each FO can communicate the decision to the electric vehicles owners.

FO scheduling

Start

DSO Congestion Analysis

Congestion Prices Indexes

Congestion?

DSO and FO final Schedule

Evs Energy Requirements

Congestion Prices equal to zero

Market Prices

Load Demand Requirements

Y

N

Fig. 2. Negotiation process

If the amount of required energy was higher than the power transformer capacity,

the DSO will determine the congestion cost and send this information to the FO. Each

FO will re-dispatch the EVs charge and discharge considering the new congestion

price. The process finishes when the congestion ceases to exist.

4 Multi-agent Simulation Platform

A multi-agent system is used to implement the negotiation process. The platform is

implemented using the JACK platform. JACK is an agent-oriented development envi-

ronment built on top of Java programming language [21].

To simulate the derived problem, four different agents were developed namely:

EV agent: An EV agent class is responsible for generating the charging

schedule of individual EVs. They communicate with the subscribed FOs.

FO agent: A FO agent class is responsible for EVs management and to the ne-

gotiation with the DSO. In short, the FO agent communicates with EV agents,

the DSO agent and the market operator agent.

DSOMarket

Operator


NotifyCongestion

ResponseConge

stion

FO_PowerSchedule

ShadowPricetoFO

MarketOperation

FO_ScheduleAD

SendPowerSchedule

FO

FO_PowerScheduleGen

eration

sends

handles

handles

handles

handles

handles

handles

handles

sendssends

sends

sends

uses

uses

uses

uses

EV

SelfPostInformation

SendEVInformation

EVSelfInformation

FO_InformationCenter

PriceCenter

FinalShadowPrice

Agent

Plan

Event

handlesposts

handles

uses

sends

handles

uses

uses

handles

handles

sends

sends

se

nds

handles

uses

handles

Fig. 3. Multi-Agent application in JACK

DSO agent: A DSO agent is responsible for the grid safety by performing load

flow calculation after obtaining the power schedules of FOs. The DSO agent

communicates with the FOs agent and the market operator agent.

Market operator agent: A market operator agent is responsible for making of

the shadow price. The market operator agent communicates with the DSO

agent and the FOs agent.

Fig. 3 shows the whole design diagram for the desired multi-agent systems in the

JACK, which is built according to the proposed solutions in this study. The diagram is

explained according to the sequence of the steps and is divided into three parts. Be-

sides, the content inside each box of this diagram (mainly the content inside the plans

of each agent) will be explained.

The interaction between agents can be explained as following:

The interaction between the EV agent and the FO agent.

o Event SelfPostInformation: Posted by the EV agent, the purpose is to

trigger the plan EVSelfInformation.

o Plan EVSelfInformation: The EV agent read the information including

initial SOC, the driving requirement of the EVs in the scheduling period,

and the bus information. After obtaining the personal information, the EV

agent sends an event named SendEVInformation to the FO agent, and the

event will be handled by the plan FO_InformationCenter.

o Plan FO_InformationCenter: The EV information will be collected

here and prepared for schedule generation.

o Plan FOPowerScheduleGeneration: An Matlab based program will be

called and be used to generate the charging power schedule. The power

schedule will be sent to the DSO agent by using the event Charg-

ingSchedule.

The interaction between the FO agent and the DSO agent

o Event SendPowerSchedule: Each FO agent sends the aggregated power

schedule to the DSO agent by this event. The event will be handled by

the DSO agent with the plan VerifyGridCongestion.

o Plan VerifyGridCongestion: The DSO agent will call the grid model

built in Matlab with the newly power schedule of the FOs and the con-

ventional loads and do the power flow calculation. The DSO agent can

fetch the value from the matlab and compare it with the capacity of the

distribution grid, such as the transformers. Then the DSO agent will send

the result to all the FO agents through the event named NotifyCongestion.

The event will be handled by the plan ReponseCongestion.

o Plan ReponseCongestion: All the FO agents get the result and check

whether congestion exists. If it is congested, all the FO agents will resort

to the market operator agent to negotiate the power capacity. Otherwise,

the FO agents are allowed to bid the energy schedule into the energy spot

market.

The interaction between the FO agent and the Market operator agent

o Event FOPowerSchedule: FO agent sends the power schedule to the

market operator agent by this event and the event will be handled by the

market operator agent with the plan MarketOperation.

o Plan MarketOperation: In the plan, the market operator agent calls the

matlab based price determination program and check whether the price is

converged each iteration. If the price is not converged, the market opera-

tor agent will send the updated shadow price to the FO agent by the event

ShadowPricetoFO. Accordingly, this event will be handled by the agent

with the plan FOScheduleAD.

o Plan FOScheduleAD: In the plan, the FO agent will reschedule the pow-

er and resend the power schedule to the market operator agent by the

some event FOPowerSchedule. If the price converged, the market opera-

tor agent will send the final shadow price to the FO agent by the event

FinalShadowPrice. The event will be handled by the FO agent with the

plan PriceCenterToEV.

o Plan PriceCenterToEV: In the plan, the FO agent uses the shadow price

to recalculate the power schedule.

Note that when the price is converged, a complete sequence of the operation

needed for grid congestion has been went through. However, the newly acceptable

power schedule of the FOs may deviate from the original plan. Therefore, we leave

the chance to the FOs and the EV owners to make a new schedule based on the infor-

mation of the first round. This explains the fact of sending the final shadow price to

the FO agent by using the event FinalShadow-Price.

5 Conclusions

The growing integration of electric vehicles in power systems introduces new chal-

lenges in the distribution networks. In many situations some congestion problems can

occur in different points of distribution networks. The most critical ones will be the

lines thermal limits and also in the HV/MV power transformers. In this paper a hier-

archical management structure is presented including the distribution system operator

(DSO) and the electric vehicles fleet operator (EV FO). The negotiation between EV

FO and DSO is discussed considering a market base negotiation in congestion situa-

tions. The implemented negotiation architecture in multi-agent platform JACK is

explained considering also the electricity market operator agent and the electric vehi-

cles owners’ agent.

Acknowledgment

The authors are grateful to the financial support of the Danish iPower project

and SOSPO project (funding from The Danish Council for Strategic Research under

grant agreement no. 11-116794).

References

[1] European Commission, “Energy Roadmap 2050,” 2011.

[2] D. M. of Climate, “DK Energy Agreement,” 2012.

[3] J. Notenboom, P. Boot, R. Koelemeijer, and J. Ros, “Climate and Energy

Roadmaps towards 2050 in north-western Europe,” 2012.

[4] European Environment Agency, “Greenhouse gas emission trends,” 2013.

[5] T. Trigg and P. Telleen, “Global EV Outlook - Understanding the Electric

Vehicle Landscape to 2020,” 2013.

[6] T. Sousa, J. Soares, Z. Vale, and H. Morais, “Day-ahead resource scheduling

in smart grids considering Vehicle-to-Grid and network constraints,” Applied

Energy, vol. 96. pp. 183–193, 2012.

[7] J. Tomić and W. Kempton, “Using fleets of electric-drive vehicles for grid

support,” J. Power Sources, vol. 168, no. 2, pp. 459–468, 2007.

[8] K. Clement-Nyns, E. Haesen, and J. Driesen, “The Impact of Charging Plug-

In Hybrid Electric Vehicles on a Residential Distribution Grid,” IEEE Trans.

Power Syst., vol. 25, no. 1, pp. 371–380, Feb. 2010.

[9] J. Hu, S. You, M. Lind, and J. Ostergaard, “Coordinated Charging of Electric

Vehicles for Congestion Prevention in the Distribution Grid,” IEEE Trans.

Smart Grid, vol. PP, no. 99, pp. 1–9, 2013.

[10] D. M. Hill, A. S. Agarwal, and F. Ayello, “Fleet operator risks for using fleets

for V2G regulation,” Energy Policy, vol. 41, pp. 221–231, 2012.

[11] W. Kempton and J. Tomić, “Vehicle-to-grid power fundamentals: Calculating

capacity and net revenue,” J. Power Sources, vol. 144, no. 1, pp. 268–279,

2005.

[12] W. Kempton and J. Tomić, “Vehicle-to-grid power implementation: From

stabilizing the grid to supporting large-scale renewable energy,” J. Power

Sources, vol. 144, no. 1, pp. 280–294, 2005.

[13] H. Morais, T. Pinto, Z. Vale, and I. Praca, “Multilevel Negotiation in Smart

Grids for VPP Management of Distributed Resources,” IEEE Intell. Syst., vol.

27, no. 6, pp. 8–16, Nov. 2012.

[14] D. Pudjianto, C. Ramsay, and G. Strbac, “Virtual power plant and system

integration of distributed energy resources,” IET Renew. Power Gener., vol.

1, no. 1, p. 10, 2007.

[15] L. Holmberg Rasmussen, C. Bang, and M. Togeby, “Managing congestion in

distribution grids - Market design consideration. How heat pumps can deliver

flexibility though well-designed markets and virtual power plant technology,”

Oct. 2012.

[16] A. Ipakchi and F. Albuyeh, “Grid of the future,” IEEE Power Energy Mag.,

vol. 7, no. 2, pp. 52–62, Mar. 2009.

[17] A. J. Wood and B. F. Wollenberg, Power Generation, Operation, and

Control. Wiley-Interscience, 1996, p. 592.

[18] T. Sousa, H. Morais, Z. Vale, P. Faria, and J. Soares, “Intelligent Energy

Resource Management Considering Vehicle-to-Grid: A Simulated Annealing

Approach,” IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 535–542, Mar. 2012.

[19] G. Heydt, “The Impact of Electric Vehicle Deployment on Load Management

Straregies,” IEEE Trans. Power Appar. Syst., vol. PAS-102, no. 5, pp. 1253–

1259, May 1983.

[20] S. Rahman and G. B. Shrestha, “An investigation into the impact of electric

vehicle load on the electric utility distribution system,” IEEE Trans. Power

Deliv., vol. 8, no. 2, pp. 591–597, Apr. 1993.

[21] N. Howden, R. Rönnquist, A. Hodgson, and A. Lucas, “Intelligent Agents -

Summary of an Agent Infrastructure.”

Appendix B

Supplementary backgroundand methods

This section includes some details that facilitate the reading in the main contentsof this thesis.

B.1 Battery modeling-Continuous

In a steady state battery equivalent circuit, by some deductions [8, 20, 28, 70],the battery current is obtained as:

I2(soc, P2) =Uoc(soc)−

√Uoc(soc)2 − 4.Rint(soc).P2

2Rint(soc)(B.1)

where P2 is the terminal power. In order to study the optimization chargingschedule for electric vehicle battery, two approaches are considered subsequently.One considers the electric vehicle battery as a pack [25,26,100], another considersthe battery as aggregated cells and usually a battery cell is investigated firstly[20, 28]. If the dynamics of the state of charge of the battery is calculated byadding kWh into the available capacity, it usually starts with the calculation of

184 Supplementary background and methods

internal power of the battery Pint which is obtained as:

Pint(soc, P2) =

{ −η.I2(soc, P2)Uoc(soc) P2 ≥ 0− 1η .I2(soc, P2)Uoc(soc) P2 < 0

(B.2)

The efficiency is a function of the battery current

η = 1 +∂η

∂c.|I2(soc, P2)|

Qmax(B.3)

where ∂η∂c is a constant reflecting the decrease in efficiency with increasing cur-

rent, Qmax is the battery capacity. Usually, linear and nonlinear approximationare used to characterize the relation between the internal power Pint and ex-ternal power P2. Using linear approximation, the internal power is assumed tobe equal to the external power. In this case all internal losses in the batteryare neglected. Using nonlinear approximation, such the study in [25] used asecond-order Taylor series expansion to derive the relation. The studies [8, 25]have shown that the difference between the two charging schedules is minor andindicates that the linear approximation is sufficient and the benefit of using anonlinear approximation does not justify the increase in computation time.

And if the dynamic of the state of charge of the battery is characterize by thecalculation of the dynamics of the electric charge, it models the battery chargingwith a first-order system which is described as:

xk+1 = xk + ∆t.I2(xk, k)

Qmax(B.4)

where xk is an actual state of charge of the cell at the step k, Qmax is the batterycapacity, I2 can be calculated the same way as in the equation B.3.

Note if the battery cell is modeled, the parameter of the variables will be scaleddown, e.g., divided by the numbers of the cells in a battery pack. For example,in order to calculate the external power P2 in the cell based model, it can beobtained by the following equation:

P2 =PBTns

(B.5)

where ns is the number of the cells, and

PBT (k) =

{−ηk.uk.Pmax k ∈ Kplug

Pdr k ∈ Kdriv(B.6)

where ηk denotes the efficiency parameter, Pmax means the maximum chargepower when the EV is connected to the grid, µk is the control variable. Now, theproblem is formulated to find the optimal control strategy u∗ = {u∗0, u∗1, ..., u∗N−1},the details are presented in the following section 3.2.2.2 or in the study [20,28].

Price control for regulating electric vehicles’ charging behavior 185

B.2 Price control for regulating electric vehi-cles’ charging behavior

In this case study, a 10kV distribution network [101] is considered where thenumber of electric vehicles is assumed to be 630 consisting 20% of the customernumbers. Normally, the distribution of the parameters α, β, γ, δ reflects theuser’s preferences, which can be obtained through survey or data feedback. Inour case, the total considered time slots are 14, i.e., the scheduling period is from16 : 00 in the afternoon to 6 : 00 in the next day’s morning. The parametersdistribution α, β, γ, δ is assumed independent, and the probability function of αis set as fα(t) = {5/39, 6/39, 7/39, 2/39, 3/39, 2/39, 1/39, 1/39, 1/39, 1/39, 1/39,2/39, 3/39, 4/39}. β, γ, δ are assumed to be uniformly distributed within [1, 20],[1, 14] and [0.01, 3] respectively. To better reflect the reality, among 630 EVs,each 20% are supposed to be charged from 4 kWh, 6 kWh, 8 kWh, 10 kWh and12 kWh to 24 kWh respectively. So the total of initial load requested from usersin the scheduling period will be X0 = 630 ∗ 20% ∗ [(24 − 4) + (24 − 6) + (24 −8) + (24− 10) + (24− 12)]kWh = 10.08 MWh. The initial requested load will be calculated by Lt =X0 ∗ fα(t). The price information p0 = [0.3876, 0.3951, 0.4734, 0.5338, 0.4943,0.4101, 0.3774, 0.3642, 0.3563, 0.3514, 0.3502, 0.3498, 0.3514, 0.3627]DKK/kWhused in A.6 is used in this case.

From the assumed value of fα(t), the probability reaches max during the timeslot 18 : 00−19 : 00, so we increase the price in this time slot to 0.9468 or 1.4202,i.e., two or three times of the original price 0.4734. The results are shown inFig. B.1 (b) and (c), respectively. For comparison, the dynamic price which isa function of the initial load Lt is also studied (Fig. B.1(d)). The relationshipis set as follows:

p(t) = 0.28 ∗ Lt + 0.2 (B.7)

Following this equation, the average price will be 0.4 DKK/kWh which isconsistent with the electricity market price.

As can be seen from Fig. B.1 (a)-(c), when the price increases at t = 3(18 :00− 19 : 00), the loads will be shifted mainly to the time slot t = 4 and flattenthe demand curve. Fig. B.1 shows that the peak of the load curve will besuppressed due to the high electricity prices resulted from equation B.2.


Figure B.1: Load profiles with price control, red curve means requested loadand blue curve represents load with price control. (a) p(t) = p0. (b) p(t) =p0,∀t 6= 3, p(3) = 0.9468. (c) p(t) = p0,∀t 6= 3, p(3) = 1.4202. (d) p(t) = F (Lt)

B.3 Congestion management principles in trans-mission systems

A

15$/MWh

B

30$/MWh

100MW 100MW

100MW

A

15$/MWh

B

30$/MWh

100MW 100MW

50MW

200MW 0MW 150MW 50MW

(a) No congestion (a) With 50MW transfer limit

Figure B.2: Two zone system.

In transmission systems, congestion management is to control the system so thattransfer limits (including thermal limits, voltage limits, and stability limits) areobserved and not violated. In the deregulated power system, the challenge ofcongestion management for the system operator is to create a set of rules thatmust be robust and should be fair as well. Robust is required because there will

Congestion management principles in transmission systems 187

be many aggressive entities seeking to exploit congestion to create market powerand increased profits for themselves at the expense of market inefficiency. Asillustrated in Fig. B.2, if there is no transfer limits between zones, all 200 MWof the load will be bought from generator A at $15/MWh, resulting a cost of$3000/h. If there is a 50MW transfer limit, then 150MWh will be bought fromA at $15/MWh and the remaining 50MWh must be bought from generatorB at $30/MWh, a total cost of $3750/h. Congestion has created a marketinefficiency of 25% of the optimal costs, even without strategic behavior by thegenerators. Congestion has created unlimited market power for generator B ifthe demand at zone B is zero price elasticity. Fairness is required because thetransmission line will probably used by many participants, as shown in Fig. B.3.If congestion happens, the tariff generated should be clear to all participantswhy the tariff has occurred. With a market approach, the form of congestionmanagement is dependent on the energy market (electricity spot market forcongestion prevention, regulating power market for buyback mechanisms).

GENCO K

GENCO E

GENCO z

GENCO HDISTCO N

DISTCO L

DISTCO Y

DISTCO M

Line limit

i j

Zone A Zone B

Figure B.3: Transmission congestion illustration: the transmission line betweenzone A and zone B will be congested due to the capacity limit.

Three forms of congestion managements have been developed for the deregulatedpower systems [102]. One form is a centralized optimization based approach,either explicitly with some form of optimal power flow program, or implicitly,depending on system operators to control congestion, found in various imple-mentations in the United Kingdom, part of the United States, and in Australiaand New Zealand. The second form is based on the use of price signals derivedfrom ex ante market resolution to deter congestion before real time operation,used in the Nordpool market area. Inevitably some congestion may still ariseand must be corrected in real time by purchasing of generation and consumptionmodifications from the system operator regulating markets, also known as buy-back. The third form seeks to control congestion by allowing and disallowingbilateral transmission, based on the effect of the transaction on the transmissionsystem, found in part of United States. Strengths and weakness of the three


techniques are further discussed in the study [102]. The techniques discussedabove can be divided into deterrent techniques, which attempt to schedule gen-eration prior to operation in such a way as to avoid congestion, and correctivetechniques, which control generation at the point of real time operation to pre-vent congestion.

Before thinking the application of the techniques. However, a question wouldnaturally arise on the similarities and differences between transmission systemand distribution network congestion as well as the solution to eliminate thecongestions on both systems, here are our analysis:

• The complexity level is different.

Congestion occurs when the scheduled energy exceeds the available trans-mission capacity in either the day-ahead or the hour-ahead market. Ifcongestion happens, the system operator will call for scheduling adjust-ment to eliminate the congestion according to the bids and offers receivedfrom producers and customers. In the transmission system level, this ispossible due to the limited producers and customers. In practical, the sys-tem operator select the accepted bids and offers and set prices to clear themarket. The decisions must maximize the economics welfare generated bythe system while satisfying the security considerations. However, this isnearly impossible to achieve in the distribution network, since thousandsof families together cause the congestions, it is quite complex and difficultto find a suitable economical principle which allocates the power capacity.

• The problem source and solution is different.

Fig. B.3 and B.4 illustrate an example of congestion happened in thedistribution network and transmission lines. In the distribution system,the overloading happens in the transformer level, it is caused by the loadbelow the transformer. Therefore, the solution is to decrease the peakload or shift the peak load. Note that distributed generation such assolar power is not considered. In the transmission system, the overloadingusually happens on the transmission lines, as illustrated in the Fig. B.3,there is a congestion on the line from bus i to bus j. To eliminate thecongestion, the GENCO and DISTCO in zone A and B can be dividedinto two groups. The first group represents the agents whose decrementaladjustments could mitigate the congestion; the second group representsthe agents whose incremental adjustment could mitigate the congestion.From this simple comparison, it seems that congestion management inthe distribution level is unidirectional, e.g., reduce the load consumptionunder transformers

• The congestion effect/cost is different.

Contract net protocols and auction approach 189

Congestion in the transmission system level can lead various marginal sys-tem price, which means the optimal resource can not utilized when thecongestion exists. The congestion in the distribution network level usu-ally effects the assets’ life and because it seldom happens currently. Alsobecause the grid is over capacity when building new lines and the typicalapproach is to plan the future capacity expansion when the capacity isfully used.

B21B22

L1L2

Line2

B1

B2

Transformer60 kV/11 kV

External grid

Line3 Line1

T1

Potential overloading Level.

L3

B3

Figure B.4: Distribution congestion illustration: the transformer and the trans-mission line 2 will be challenged due to the increasing load below it.

B.4 Contract net protocols and auction approach

General equilibrium market mechanisms use global prices, and the price is usu-ally coordinated by a centralized mediator. Contract net protocols [103], how-ever, defines that individual nodes are not designated a priori as managers orcontractors, these are only roles, and any node can take on either role dynam-ically during the course of problem solving. In his article [103], Smith definedthat a manager is responsible for monitoring the execution of a task and pro-cessing the results of its execution. A contractor is responsible for the actualexecution of the task. Although this is an effective way for reach an consensuswhen facing the conflicting resources, it is not suitable for the present problemsfrom the system operator’s perspective.

Auction can be designed as either single-sided auctions such as the case in the


regulation market or two sided auctions such as the case in the spot market.In the single-sided auction, offers are ranked in increasing order and acceptedbeginning with the least expensive and continuing until the demand is satisfied.The uniform price is then set equal to either the last accepted offer (LAO) orthe first rejected offer (FRO). In two-sided auctions, offers and bids are rankedas above, and an equal quantity of each is accepted, beginning with the highestbids and lowest offers, until supply or demand is exhausted or the offer priceexceeds the bid price. It is noted that the uniform price auction mechanism isusually combined with optimal power flow calculation [104], which mean eitherDSO/Market operator will implement a lot of calculations for distribution gridcongestion management when using the uniform price auction mechanism.

B.5 Decomposition methods and subgradient method

This subsection provides a tutorial on the method of combining the subgradientmethod with dual decomposition techniques, as they are the mathematical tech-niques which support the development of the market based control. We firstlyintroduce the definition of subgradient. Subgradient, alternatively named sub-derivative, subdifferential generalizes the derivative to functions which are notdifferentiable. Let f : I → R be a real-valued convex function defined on anopen interval of the real line. Such a function need not be differentiable atall points. For example, the absolute function f(x) = |x| is nondifferentiablewhen x = 0. However, as seen in the figure B.5, for any x0 in the domainof the function one can draw a line which goes through the point (x0, f(x0))and which is everywhere either touching or below the graph of f . The slope ofsuch a line is called a subderivative (because the line is under the graph of f)and that a subgradient of f at x0 is any vector g that satisfied the inequalityf(x)− f(x0) ≥ gT (x− x0) for all x. When f is differentiable, the only possiblechoice for gT is ∆f(x), and the subgradient method then reduces to the gradientmethod. The set of all subgradients at x0 is called the subdifferential of f at x0,denoted ∂f(x). So the condition that a subgradient of f at x0 can be writtengT ∈ ∂f(x)

To introduce the subgradient method, we start with an unconstrained case,where the goal is to minimize f : Rn → R, which is convex. The subgradientmethod uses the simple iteration

x(k+1) = x(k) − αkg(k). (B.8)

Where x(k) is the kth iterate, g(k) is any subgradient of f at x(k), and αk is thekth step size. Thus, at each iteration of the subgradient method, we take a step

Decomposition methods and subgradient method 191

Xo

Figure B.5: A convex function (blue) and ”subtangent lines” (red) at x0.

in the direction of a negative subgradient. Further discussions such as step sizerules, convergence results are presented in [105]

As discussed in [105, 106], subgradient methods are often applied to large-scaleproblems with decomposition techniques. Such decomposition methods oftenallow a simple distributed method for a problem. The original primary moti-vation for decomposition methods was to solve very large problems that werebeyond the reach of standard techniques, possibly using multiple processors.It is still a good reason to use decomposition method for some problems. Butother reasons are emerging as equally (or more) important such as decompositionmethods yield decentralized solution methods in many cases. Two decomposi-tion methods are presented in [105, 106], i.e., primal decomposition and dualdecomposition. We will briefly introduce the dual decomposition method sincethe dual variables (shadow prices or prices) are manipulated to solve the globalproblem. To introduce the dual decomposition method, we considers a simpleexample where the two subproblems are coupled via constraints that involveboth sets of variables

minimizef1(x1) + f2(x2)


subject tox1 ∈ ζ1, x2 ∈ ζ2, h1(x1) + h2(x2) ≤ 0 (B.9)

where ζ1 and ζ2 are the feasible sets of the subproblems, presumably describedby linear equalities and convex inequalities. The functions h1 : Rn → Rp andh2 : Rn → Rp have components that are convex. The subproblems are coupledvia the p constraint that involve both x1 and x2. By using dual decomposition,the problem B.9 is transformed into a partial Lagrangian problem,

L(x1, x2, λ) = f1(x1) + f2(x2) + λT (h1(x1) + h2(x2))

= (f1(x1) + λTh1(x1)) + (f2(x2) + λTh2(x2))

which is separable, therefore, the problem can be minimized over x1 and x2separately, given the dual variable λ, to find g(λ) = g1(λ) + g2(λ). The g1(λ) isfound by solving the subproblem

minimizef1(x1) + λTh1(x1))

subject tox1 ∈ ζ1 (B.10)

and the g2(λ) is found by solving the subproblem

minimizef1(x2) + λTh1(x2))

subject tox2 ∈ ζ2 (B.11)

To find a subgradient of g, the master problem objective is to get solution x1and x2, respectively. A subgradient of −g is then h1(x1)+h2(x2). It is proposedin [105,106] that a simple algorithm based on subgradient method can be usedto update λ.

Repeat

Solve the subproblems:

Solve (B.10) to find an optimal x1.

Solve (B.11) to find an optimal x2.

Update dual variables (prices) until the prices converge:

λ = (λ+ αk(h1(x1) + h2(x2)))+

Market based control-Negotiation types 193

B.6 Market based control-Negotiation types

It is discussed in [107, 108] that the negotiations can be characterized as eitherstrident antagonist or cooperative antagonist. The former is characterized bycompletely distrustful and malevolent (towards one another) parties, as wouldbe the case when authorities negotiate with kidnappers or airline hijackers. Thelatter is characterized by entirely self-interested and disputing parties but onesthat recognize and abide by whatever rules exist. A third type of negotiationis called fully cooperative, where the parties have different needs, values, andopinions, but they share information, expect total honesty, perform no strategicposturing, and think of themselves as a cohesive entity with intention to arriveat the best decision for the entity, as would be the case for a happily marriedcouple. We are interested here in the two different levels of cooperative ne-gotiations, since they better typify the various types of power system decisionproblems. For example, a negotiation involving all smart appliances about theuse of a certain amount of electricity at a smart home is a good example of fullycooperative negotiation, since the fundamental aim is to optimize the operationof the electricity at residential level. On a higher level, two commercial aggre-gators negotiating on a common transmission line could be a good example ofcooperative antagonists.

B.7 Alternating direction method of multipliers(ADMM)

The alternating direction method of multipliers are used here to devise the pricemechanism which can be used to solve the grid congestions. We firstly dividea distribution network into several sub-networks, each network is associatedwith inputs such as the local renewable energy, and the power transfer line andoutputs such as the load, the power transfer line. We start the problem analysisby formulating it into an optimal power flow (OPF) based problem. Typically,the OPF can be solved centrally. However, in the distribution network, it mightbe infeasible considering the number of small units, the fluctuating generationand load profiles. Instead, we turn our solution to the distributed techniques,i.e., the ADMM solution [109]. The proposed method is illustrated by thefollowing pseud code:

Step 1: DSO designs the sub-network systems and specifies an operator for eachsub-network.

Step 2: Price coordination


Figure B.6: A illustrative distribution network.

Parameters initialization.

Iterations:

(1) EV fleet operator calculates the power and phase schedules of EVs and passesit to the sub-network operator (e.g., a function of DSO).

(2) DSO line operator calculates the power and phase schedule of the transferlines and passes it to the sub-network operator.

(3) DSO sub network operator computes the new average power imbalances andphase residual, updates its dual variables which reflects the imbalances, normally,it is called shadow price. The shadow price will be broadcasted to the associatedactors.

(4) Termination condition checking, if the net power imbalance and phase in-consistence across all sub networks of the system is less than a small value, thenit is balanced, otherwise, go to number one in step 2.

We simulate the proposed method in a distribution network. Fig. B.6 depictsthe relevant loads and lines inside the network. Several EV fleet operators areregarded as the aggregated load and numbered as load 1, 2, 3, 4. Five sub-networks are defined. We consider the power line’s power constraints as wellthe voltage limitations.

www.elektro.dtu.dk

Department of Electrical EngineeringCentre for Electric Power and EnergyTechnical University of DenmarkØrsteds PladsElektrovej 348DK-2800 Kgs. LyngbyDenmark

Tel: (+45) 45 25 38 00Fax: (+45) 45 93 16 34E-mail: [email protected]

Control strategies for power distribution networks with ... · EVs into the power distribution...

Documents

Transcript of Control strategies for power distribution networks with ... · EVs into the power distribution...